CN106407196A - Semantic analysis intelligent instruction robot applied to logistics management software - Google Patents

Abstract

The invention refers the conventional natural language analysis technology, and discloses a language analysis and instruction analysis algorithm based on large logistics management software. In one aspect, instruction information of an instruction of a natural language input by a user can be extracted through a semantic analysis engine, wherein the instruction information includes key information such as an instruction operation target, an operation command and parameters; in the other aspect, an execution demand of the user and an actual resource condition of a system is combined through an instruction analysis engine, and an executable scheme including an AJAX format command and description information is fed back to the user; and a semantic analysis intelligent instruction robot applied to logistics management software is established by combination of the two aspects, the user can directly input a natural language type instruction to directly operate the logistics management software, and the fuzziness and ambiguity of the natural language can be supported to some extent; the problem that the conventional management software is complex in interface, and an operator is difficult to operate the conventional management software can be solved; man-machine interaction is humanized and simple; and the operation efficiency is high.

Description

Semantic analysis intelligent instruction robot applied to logistics management software

The technical field is as follows:

the invention is based on a logistics management software platform, and by using the existing natural language analysis technology for reference, a series of algorithms and processes for language analysis and instruction analysis applied to the logistics management software are independently developed, the commands of the natural language type input by the user can be analyzed, the ambiguity and the ambiguity of the natural language input by the user are supported to a certain extent, and the intelligent instruction type robot capable of directly operating the logistics management software is constructed. The user can directly operate the logistics management software through the input language. The technical platform relied by the invention is comprehensive logistics management software, aiming at the common problems of complex interface, complex operation and difficult hands of the user of large-scale management software, the intelligent instruction type robot created by the invention can lead the user to directly operate the system through simple instructions closer to daily language, the human-computer interaction is more humanized, the hands are faster, and the operation efficiency is favorably improved.

In the intelligent robot designed by the invention, an operator can directly input a natural language type command: such as: "the party of creating the manifest is Pebax Cola, the goods is 500kg Cola" or "create financial voucher, the object is Zhangai, receival is 5000 soon" or "search for the delivery order of the latest month's Chengdu site" and so on. The language input by the user is automatically analyzed into one or more commands which can be understood and executed by the system through a semantic analysis engine and an instruction analysis engine of the intelligent robot, each command corresponds to descriptive explanatory characters and is fed back to a client in an input interface, the client can select the command according to actual needs and click to execute, and the operation interface is shown as an attached drawing 11.

On one hand, the invention uses natural language analysis technology and algorithm for reference, and is based on the application range of the invention: the logistics management software autonomously develops a set of intelligent semantic analysis algorithm, analyzes input sentences of a user into instruction resources to be executed, including the resources of commands, targets, parameters and the like of executing instructions, by an instruction resource library provided by a physical management software platform, and can support the ambiguity and the ambiguity of natural language to a certain extent. On the other hand, the system can analyze a series of executable schemes for the customer to select according to the requirement input by the user and the resources of the system, wherein the executable schemes comprise description information of executable instructions and execution instruction links.

Background art:

the background technology on which the present invention is based has mainly two directions: the system comprises a large logistics management software platform and the existing natural language analysis technology.

The large-scale logistics management software based on the invention is comprehensive management software for the logistics transportation industry, covers a plurality of aspects of logistics enterprise operation processes such as transportation management, warehousing management, customer relationship, financial management, office automation and the like, has the advantage of powerful functions on one hand, but also has common diseases of the large-scale management software: the process is complex: the user is difficult to use and the resource searching efficiency is low. The technical platform of the software is developed based on a J2EE platform, background core code resources are mainly composed of JAVA codes, the software platform realizes a B/S (browser/server) framework, and a user does not need to specially install a client but directly operates a system in a browser.

On the other hand: the existing natural semantic analysis technology is gradually mature, and a plurality of intelligent semantic analysis robots based on the semantic analysis technology, such as a Turing robot, appear. The method can be based on natural language processing, realizes self-learning and self-defined knowledge base expansion, and by far, Turing robot platforms accumulate nearly hundred million corpora and GB-level basic data, and the identification accuracy reaches 85% through determination of an authoritative institution. However, in the current market, the application based on the semantic analysis technology and the extended application based on the semantic analysis robot also focus on language processing, such as the application in the aspects of a translation robot, a chat tool and a knowledge question and answer tool, and the function of language intelligent analysis can not be really played.

The invention combines the natural language analysis technology and the logistics management software, and constructs a more efficient and more humanized operation interface: namely, the semantic analysis intelligent instruction robot of the logistics management software can be directly operated, and the structural flow of the whole set of system is shown in fig. 1. The intelligent instruction robot provides a simple interface similar to common chat software, as shown in the attached figure 11, a user does not need to operate in a complex system interface, and only needs to input language type instructions, such as ' create waybill XXX ', inquire client information XXX ' and the like, in the interface as simple as a chat tool. The intelligent robot analyzes the possible operation intention of the user according to the user input and the actual resource condition of the system, and feeds back various feasible executable schemes to the client in the same input interface. Each set of executable schemes has complete description information, links for executing instructions and a series of parameters. And the user selects the operation scheme required by the user to execute according to the information fed back by the system.

The human-machine interface of the workflow management software operated by the intelligent command robot by the operator is a simple interface similar to chat software, as shown in the attached figure 11. In such an interface, the user gives commands like a system to get feedback from the system and execute the commands like a daily chat. The invention simplifies the original management software interface with complex flow and complicated operation into a simple chat interface, and the operation of the user is changed from the traditional mechanical operation mode into an operation language closer to human language, so that the man-machine interaction is simpler and more humanized.

The operation flow of the intelligent robot is as follows: as shown in figure 11 of the accompanying drawings: the user clicks an open operation interface to input a command: such as: "build manifest shipper is Pebax Cola and shipper is 500kg Cola". The robot judges sentences which may need to be executed by the user according to the instruction input by the user and the actual resource condition of the system, feeds the sentences back to the user, and displays a plurality of executable schemes on the interface of the chat tool: such as: and 1, creating a waybill, wherein a shipper is an enterprise client: [ Baishi Cola Chengdu division ] with the names of goods: [ Cola ], weights [ 500 ] kg "and" 2. build the waybill, the shipper is the business customer: [ Baishi Cola Rhinoceros plant ] with the goods name: [ Cola ], weight [ 500 ] kg ", etc., and is followed by a link for a click operation.

The user selects a certain instruction according to actual needs, clicks the instruction, and directly enters the waybill creation page, and the page has set parameters according to information contained in the user instruction, for example, the shipper is set as "Baishi Cola Chengdu Subdivie", the name of the goods is "Cola", and the weight is 500kg, as shown in fig. 12 of the attached drawings.

It can be seen from this that: the semantic analysis robot of the logistics management software carries out semantic analysis and instruction analysis on user instructions, so that the instructions of user language types can directly operate a system, the logistics software is more humanized to use, more close to the daily use habits of people, and more easy to use. The most suitable use scenario of the invention is as follows: firstly, the user who is unfamiliar with system function and resource distribution can help them to enter the door to the system more quickly, can also find the resource that needs more effectively, secondly the more loaded down with trivial details function of operation, for example waybill is established, need set up information such as delivery side, goods name, weight, need switch back and forth at several pages just can accomplish, can once only dispose required resource through intelligent robot, improved the efficiency of execution.

Fundamental principles, procedures and difficulties of technical implementation of the system:

compared with the operation mode in the traditional interface, the operation mode of the command of the nature of the natural language input by the user has the following difficulty:

the first difficulty is represented by the ambiguity and inaccuracy of the input instructions: the ambiguity is determined by the nature of the user language, the daily used language is not 100% accurate, the language input by the user language cannot be 100% matched with the instruction resource capable of running in the system, even if the user is familiar with the system resource, the input language instruction is ensured to have definite directivity, the language and the instruction capable of being executed by the system have certain deviation, in other words, under the ordinary condition, the user completely initiates a command to the system according to the daily speaking habit, and the ambiguity is larger. As described in the previous paragraphs, in the actual operation process, the user may feed back various executable schemes to the user according to the input of the user language instruction and according to a certain fuzzy matching degree, and the user may install the instruction input at will, and the system may often feed back a plurality of executable instruction schemes for the user to select and execute.

Secondly, inaccuracy is brought by semantic robot operation, when a command is submitted in a traditional system operation interface, firstly, the command has accurate parameters, and in the interface of the intelligent robot, the command input by a user language cannot have 100% of completely accurate parameters; for example, in a conventional interface, when a certain waybill is selected for editing or viewing, hidden parameters submitted to a background by a page include a UUID of the waybill, which is usually a 16-bit or 32-bit unique code, and can uniquely determine documents such as the waybill. In the interface of the intelligent robot, the user can only input fuzzy and inaccurate parameters close to daily language habits, and cannot input the parameters as accurate as UUIDs.

Secondly, when the traditional operation interface is operated, different pages represent different service scenes, the page also has hidden context information, for example, all operations are performed on the waybill inside the waybill page, and therefore the system can be automatically mapped into an operation target resource of a system background waybill by operating inside the waybill page.

And further: each operating button or action in the traditional interface is also completely in one-to-one correspondence with a certain command in the background, for example, a certain waybill is edited, a background 'edit' command is executed, a 'waybill new-creation' button is clicked, and a background 'new-creation' command is executed. In the interface of the general intelligent instruction robot, the context information brought by traditional pages, buttons and other UI elements is not available, and the only information source is the analysis of the user language.

Handling of inaccurate user instructions: firstly, an algorithm supporting ambiguity is built in the existing semantic analysis engine, keyword matching can be performed according to the matching degree set on each resource node, such as the matching degree of 80%, and some ambiguity and ambiguity parameters are also preset in the semantic analysis engine and finally matched into a plurality of accurate system instruction resources. The second is an instruction analysis engine (refer to fig. 1: system structure flow) behind the system, which performs related processing for the ambiguous instructions: fuzzy instruction resources are set in the instruction resource library, the resources comprise fuzzy execution targets, commands and parameter resources, the fuzzy instruction resources are not finally executable resources, and the fuzzy instruction resources are further matched with accurate and finally executable data resources according to a certain algorithm. Such as: the 'delivery party' is an ambiguous instruction parameter resource in the system, and when a user inputs 'delivery party' information, the 'delivery party' can be matched with the 'individual delivery party' and 'enterprise delivery party' which are really executed according to a certain algorithm.

The second difficulty is how to construct the instructions and accurate description information that the final system can execute through the commands of natural language nature input by the user.

The user inputs instructions of natural language nature, and finally needs to feed back to the user a plurality of feasible execution schemes, including page links with clickable execution instructions and detailed description information, and the step of constructing executable instructions includes the step of constructing executable commands and related parameters and accurate description information of each execution scheme.

The logistics management software based on the invention is a B/S (browser/server) architecture, the final command execution is realized by submitting an AJAX format instruction to a background, and the content submitted by the AJAX instruction comprises a command execution address and command parameters.

Composition of AJAX instruction resource addresses: the former part is the general address of the server and the project, the rest can be constructed by two parts of an instruction target and a command, for example, when the waybill is created, the command address to be submitted is:http：//127.0.0.1：8080/torstein/bookingNote/create.html. Wherein,http：//127.0.0.1：8080/torsteinthe part is the public address of the server and the engineering project, and the part shared by all the resources of the system. "bookingNote" is a resource address of an instruction target, and indicates that a target of current instruction execution is "manifest" (bookingNote); "create" is a resource address of an instruction command statement, and indicates that the executed command is "create". From the above analysis, the address submitted by the instruction execution can be combined by the execution target and the resource address of the execution command.

Meanwhile, certain parameter information is also required for the execution of the instruction, including parameter names and parameter values. As in the previous example of waybill creation, when creating waybill by AJAX command in the system, if the shipper is designated as an existing enterprise client of the system [ hectares coke capital division ], the parameter information to be submitted during execution is: sender corporation: "Pascal Cola Chengdu sub", i.e., the parameter name is "sender corporation", and the corresponding parameter value is "Pascal Cola Chengdu sub".

From the above analysis, it can be seen that the commands and parameters that can be executed by the system are constructed, and three necessary factors must be obtained through the language analysis of the user: 1. operation target, 2. operation command, 3 command parameter. The three factors are used as instruction resources to be stored and managed according to a tree structure, the instruction resources stored in the instruction resource library of the system are called secondary response instruction resources, the resources are controlled by an instruction analysis engine based on physical management software, and the resources are matched and mapped again in the instruction analysis process. The structural diagram of "secondary response instruction resource" refers to fig. 2. The root directory of the tree structure is an operation target resource, the leaf nodes are parameters of executable commands, and each level of instruction resources stores address elements corresponding to the level of resources. After finding the finally matched secondary response instruction resource, the instruction analysis engine constructs the address resource of the whole operable command according to the sequence from the root node (operation target) to the leaf node (command parameter) and the address element. The description information of the executable instruction is similar to the process of command construction, and is also formed by an operation target, a command and an instruction parameter in sequence, the secondary response instruction resource not only comprises the address factor of the executable command, but also comprises the description information, and after the instruction analysis engine finishes the matching of the instruction resource, the description information capable of executing the instruction is constructed according to the sequence from the root node (the operation target) to the leaf node (the command parameter).

The third difficulty is the semantic analysis technique of natural language: namely how to extract key information such as basic operation targets, commands and parameters from the natural language input by the user. According to the above description, the instruction analysis engine needs to extract the target, command and parameter information from the user language as input signals respectively, and then match and analyze again in the instruction analysis engine and the system resources.

The problem is that how to implement the natural language analysis technology in the research process experiences more difficulties, and the problem is mainly focused on the implementation direction of the semantic analysis technology in the research process: firstly, a language analysis technology is realized based on the existing intelligent semantic analysis robot, such as a Turing robot; and the second is an independent development language analysis algorithm which is specially used for the logistics management software.

Finally, the advantages and the disadvantages of the two directions are analyzed, the application scenes of the system are combined, the existing semantic analysis technology is used for reference, a semantic analysis engine based on an instruction resource library is automatically developed, and the system is specially used for the logistics management software.

This is chosen for the following reasons:

1. the semantic analysis related to the invention has limited application range and vocabulary: the system is only limited to be used by logistics management software, instruction analysis capable of operating a system is performed, the system is not common chatting software, a vocabulary library for semantic analysis is limited, manual maintenance can be performed, and the massive vocabulary library of the existing intelligent semantic analysis robot is not required to be used.

2. The processing context is simple, and the sentence length to be analyzed at one time is limited: the length of the analysis language is limited to one sentence, only limited verbs and nouns need to be analyzed, auxiliary words, adjectives and the like do not need to be analyzed, and the context before and after the analysis does not need to be considered, so that no complex language environment needs to be processed, complex ambiguity and ambiguity in the natural language need not to be processed by using a complex semantic analysis algorithm, and the processing of the ambiguity and ambiguity of the language can be realized by self-designing a semantic analysis engine and a back-end instruction analysis engine. If words which cannot be analyzed by the vocabulary library are encountered, the words can be directly treated as invalid instructions.

3. The existing semantic analysis robot has single function of the interface provided at present: for example, the turing robot interface which was examined earlier cannot provide support for variables, so that the transmission of instruction parameters in a logistics management software scene cannot be realized, and the support for the subsequent instruction analysis process cannot be realized.

The invention content is as follows:

in the invention, the whole process of processing the user command is actually a process of generating an executable scheme after a series of analysis processing through a natural language instruction input by a user.

In the logistics management system, the executable scheme consists of commands in AJAX format, parameters (AJAX instruction links capable of being executed) and descriptive language.

The whole flow of user command processing is shown in fig. 1, and can be divided into three stages: semantic analysis, instruction analysis, and instruction resource combination. The semantic analysis stage is executed by a semantic analysis engine, and natural language of a user is analyzed into 'primary response instructions' with structural information, such as most basic targets, commands, parameter information and the like according to system instruction resources. Instruction analysis is performed by an instruction analysis engine. The 'primary response instruction' is further analyzed and converted into the address and parameter set of the AJAX command which can be executed really.

A semantic analysis stage flow: analyzing a natural language command input by a user, and extracting key information of three aspects: target of an instruction, command action of an instruction, and parameter information.

The semantic analysis engine performs semantic analysis as follows:

the first step is as follows: first, the engine loads the instruction resource library, and the data structure of the instruction resource library is shown in fig. 2: the data structure exhibits a tree structure and is divided into "primary response resource" and "secondary response resource" portions. The "secondary response resource" part is a resource main body, and includes nodes having a hierarchical structure relationship from a root node "instruction target", "command", to a leaf node "parameter", and the like, and these nodes in the "secondary response resource part" are referred to as "resource nodes".

And each level of resource nodes is hung with a primary response node which is used for matching with the input language of the user according to the set matching degree and the matching content in the semantic analysis process. As can be seen from fig. 2, all "resource nodes" will include "primary response nodes" in the instruction resource pool, which are collectively referred to as "primary response resource portion".

The second step is that: and (3) implementing keyword matching: according to the matching content (content) and the matching degree (matchDegree) set by each level of 'primary response resource' nodes, statements input by a user are sequentially matched according to a character string forward matching algorithm provided by an engine, when a certain keyword is successfully matched, the successful 'primary response resource' is added into a return parameter, and finally the return parameter is a list consisting of a series of 'primary response resource' node instances, called as 'primary response instruction', and is input into a subsequent instruction analysis engine.

Matching the semantic matching sequence according to the sequence of the target resource, the command resource and the parameter resource: firstly, trying to match the target resource, and after the target resource is successfully matched, trying to match the child nodes of the command resource below the target resource according to the tree structure of the response resource, as shown in step 1 and step 2 in fig. 3. The legal "primary response instruction" must be successfully matched from the user's language to the instruction target information and the command information. If the target resource and the command resource cannot be successfully matched in the input statement of the user, the matching fails, and the user inputs an illegal command statement.

Matching process of target resource and command resource: the matching process of the two resources is relatively simple: after observation and statistics of user input sentences, the target resources and commands are generally in a single structure and are displayed in plain text in user instructions.

In the semantic analysis process, matching is directly carried out on sentences input by a user according to matching content (content) and matching degree (matchDegreee) by utilizing a character string forward matching algorithm provided by an engine, and once matching with keywords is found to be successful, the two corresponding 'primary response resource' nodes are added into a returned parameter 'primary response instruction'.

FIG. 3 shows a matching process for a target resource and a command resource. When the user inputs the sentence as: when the party for creating the shipping bill and sending the goods is the Pebax Cola and the goods is 500kg Cola, the matching process is as follows:

step 1: the target resource is first matched. According to the instruction resource library, the target resource: matching contents of the primary response resource node hung below the freight note BookingNote comprise keywords of the freight note, and if the target resource is successfully matched, the target resource is added into a return parameter: in the "primary response instruction". Similarly, step 2, in the target resource: matching contents of a node of a 'primary response resource' governed by 'Create' in a command resource below 'waybill BookingNote' comprise a key word of 'Create', and if the command resource is successfully matched, adding the matched content into a return parameter: in the "primary response instruction".

And (3) processing semantic matching ambiguity of the target resource and the command resource: if there are many target resources and command resources that can be matched according to the matching degree and matching content set by the system, many executable schemes are finally generated for the user to select.

Matching process of parameter resources:

the parameter resources and the structure of the parameter resources are complex, so that the matching logic and the matching process are complex. The data structure of the parameter resource not only comprises parameter names, but also comprises parameter values, and some parameters present a composite structure: in addition to the parameter values themselves, with unit and operator signs (e.g., greater than, less than, etc.). In the language instruction input by the user, only one target resource and one command resource are provided (the current system only supports a single command statement and can only execute a single operation and a single target), and the plain text display in the statement can be realized through direct keyword matching. In the input statement of the user, the input parameters can be multiple, each parameter does not necessarily have obvious keywords to match, each parameter has complex parameter values, units and operation symbol information, and different parameters have related relations.

In this case, the semantic analysis engine extracts parameters in several ways: in the first case: the parameters of the single structure can be matched with the parameter values in a keyword direct matching mode, such as' a delivery party: the parameters of the Baishi Cola type are matched and directly extracted by keywords in a mode similar to an instruction target and a command, and can be accurately obtained by: parameter name: "sender", parameter values: "Baishi Cola".

In the second case, the parameters comprise a composite structure: we found by statistics: the parameters with a composite structure mainly have two additional structures of units and operators. For parameters of a single structure, such as "shipper" and "consignee" parameters. In the resource data structure, the parameter resource node is a leaf node, and then there is no child node below. For the parameter resource with unit and operator, the instruction resource library has two sub-nodes below the parameter node: "parameter unit node" and "parameter operator node". As shown in FIG. 4, the two accessory nodes are also respectively hung with a "primary response node" for semantic analysis and matching.

Such as: "cargo weight": the data structure of this parameter resource with parameter units in the resource library is shown in fig. 5. The parameter resource node is named as 'cargo weight', and a plurality of related parameter unit nodes are hung below, such as: kilogram, gram and ton. Each parameter unit node includes a "primary response node", for example, a "primary response node" hung by a unit node of "KG" includes matching content of "KG, KG", and when the language input by the user includes such a keyword, the matching content can be matched to the unit node of "KG".

In the third case, the parameter itself includes not only the complex structure itself, but also other subdivision parameters, and such "parameter" is essentially a general term made up of several related parameters. For example, in the previous example: "cargo: the parameter of 500kg coke and goods is actually a set of a series of sub-parameters related to goods information, and actually contains more detailed hidden parameters such as weight of goods, volume of goods, name of goods, and the like. If the parameters obtained according to the direct matching method-goods: "500 kg Cola" is not directly matched to "goods name" nor to "weight of goods". And performing secondary matching, further segmenting words and further matching the parameter values obtained by the primary matching according to certain logic, and obtaining subdivided parameters which really play a role: "goods name: cola, weight of cargo: 500 kg.

The logic of how to implement such subdivision, quadratic matching parameters: on the node of the primary response resource, by setting the field: and (3) nextmappingproxycass (secondary matching agent class), realizing the logic of secondary matching, and obtaining the 'primary response resource' node which really plays a role next time. When the class is set to be empty on the node, secondary matching is not needed, and after the matching is successful, the instance of the node is directly used as a 'primary response resource' node which is successfully matched to return; if the class setting is not null, after the node matching is successful, the semantic analysis engine needs to instantiate an instance of the nextmappingproxycalslass (secondary matching agent class), and implement the logic of secondary matching by using the method maptoenextcommand provided by the class, and return a new return value: i.e. there are several instances of the list of "primary response resource" nodes.

The foregoing "cargo weight" example is also achieved by this "second matching" process. When the user inputs "goods: when 500kg coke is used, the keyword of "cargo weight" or "weight" is not mentioned in the input sentence of the user, and according to the "cargo weight" parameter resource structure shown in fig. 5, the parameter node of "cargo weight" itself cannot be matched, but the matching needs to be realized according to the keyword provided by the following parameter unit node "kg", that is, the character "kg" input by the user is matched to the parameter unit node, and then the "secondary matching agent class" of the unit node is started, as shown in the example code of fig. 6, according to the logic provided by the agent class, the keyword of "500" is reversely matched to the parameter node of "cargo weight", and 500 is used as the parameter value of the parameter of "cargo weight".

Fig. 6 shows an example of this class of secondary matching agents, which is defined on the "kg" node, i.e., the "cargo weight" parameter node, and the following "kg" node in weight units, the core logic being represented by the code in the red box: once the node is successfully matched, that is, contents which can be matched with the node with the weight unit of "kg" are found in the input sentence of the user, the secondary matching agent class is started by the semantic analysis engine, and the method is entered: maptoenextcommand.

The core logic of the method is as follows: first, using the product cargo template class landefproducttestemployprox provided by the semantic analysis engine framework, return the unit of cargo weight and write to the return value "primary response resource": list < LanuageFirstCommand > result. In addition, according to the sentence subRawSentence extracted from the previous node "cargo cargoWeightvalue", if the keyword cargoName of the cargo weight and the key value cargoName of the cargo name can be extracted from the product cargo template class according to the preset extraction mode, and depending on the context interface exposed by the langue resource order command repository, the node for the cargo weight and the node for the cargo name are respectively extracted from the cargo weight "kg" node, and the related values are given to their "primary response node" and added to the return value result.

The matching and processing method for useless and meaningless characters in an input command comprises the following steps:

in the foregoing example of semantic analysis, we can see some words playing a connecting role, such as "yes" in "cola for goods", and all spaces, punctuations, etc., which do not have any practical role in command operation, and need to be filtered out in the semantic analysis, and this time, the words are also important word segmentation identifiers, which indicate that some keywords, such as parameters, commands, etc., match up with the end mark. In the system, the 'primary response nodes' are adopted for matching for the identification of the meaningless and useless words, and keyword matching is carried out according to the matching content and the matching degree defined on the 'primary response nodes'. But with the difference that: the 'primary response node' of the nonsense word is not bound to a certain 'resource node', but presents a 'free state', and once the characters are matched with the 'primary response node' of the 'free state', the semantic analysis engine considers that the sentence corresponding to the node is the nonsense word and needs to be filtered out in the processing process.

In summary, the semantic analysis engine analyzes the resources of the instruction resource library, and performs matching based on the "primary response resource" node, and finally obtains a list composed of a series of successfully matched "primary response resource" node instances, which is called as a "primary response instruction", and sends the list to the subsequent instruction analysis engine, and the next instruction analysis is performed.

And the instruction analysis engine is used for further analyzing the 'primary response instruction' of the list structure into a secondary response instruction of a composite structure, wherein the information contained in the secondary response instruction meets the AJAX execution format on one hand, and also meets the requirement of feeding back enough and readable information to a user for screening.

The secondary response instruction adopting the composite model structure contains more abundant information, as shown in fig. 2, the secondary response instruction resource presents a tree structure taking a "target (target) resource" as a root node, a child node of a command (command) is below the target node, which represents all executable commands possibly contained below one target resource, and the command and the target are of a composite structure and both contain a resource number id, a display text label and a command link url.

And (3) an instruction analysis process:

the data structure of the instruction resource library is shown in fig. 2, all executable instruction resources in the system are defined through a tree structure, and a root node of the model is a 'target resource'. The logistics management system based on the instruction analysis engine and each business model adopted by the underlying platform are target resources. Such as creating a new waybill, searching for waybill, or creating a new financial account. The business model here: the 'waybill' or 'financial account' is a business model of the logistics management system and is also a target resource of the instruction. The "command resource" is a child node below the "target resource", and one target resource may include a plurality of command resources, for example, the "waybill" includes a plurality of command resources below: such as new creation, editing, auditing, etc.

The process of executing instruction analysis by the instruction analysis engine comprises two mapping processes: the first time is to map the target and command resources of the "secondary response instruction," which may also be referred to as "command mapping. And directly returning the corresponding secondary response instruction node according to the instruction target node and the command node in the primary response instruction and the structure in the instruction resource library.

The second mapping, namely the "parameter mapping", is more complex than the first mapping, and the mapping process is performed by a control field on the parameter resource: the "mapping decision class" decision, which represents the mapping logic for each input parameter, needs to be defined if the parameter needs the relevant logic to be directed to other parameters during the mapping process.

When the mapping decision class is set to be null, a direct mapping mode is adopted, namely the currently obtained parameter resources are directly used as input parameters of the instruction resource combination of the next stage; if the setting of the mapping decision class is not null, mapping is required to be carried out on one or even a plurality of new parameters according to the definition of the decision class, if the current parameters are required to be mapped on a plurality of parameters, a plurality of secondary instruction responses are formed, and finally, a plurality of executable instructions are returned to the client after the subsequent instruction resource combination. If the mapping decision class is empty, a mapping mode of direct matching is adopted, that is, the name of the parameter in the initial response instruction is directly matched with all the parameters under the command resource obtained by the first matching, and after the matching is successful, the command resource is directly sent to the third stage to be combined, and fig. 8 describes a control flow of the two-time mapping of the command response resource in the command analysis process.

In the actual operation process, the instruction analysis engine dynamically instantiates according to the mapping decision class set by the parameter resource and performs mapping logic control.

The flow of the "mapping decision class" control mapping is illustrated: assume that when the "waybill creation" command is executed, the parameters input by the user in the primary response instruction are: the delivery side (sender) is "Petasy Cola": in the background system, the actual resources corresponding to the delivery party are client resources of two types, namely 'personal client' and 'enterprise client', and when the specific client resource type cannot be judged by a command input by a user, the first matching of the secondary response resources is as follows: the general type of "sender" is not the final parameter that can be directly executed, and needs to be mapped to the real executable "personal client" and "enterprise client" according to a certain logic.

The parameter resource of the "sender" sets a "mapping decision class" in the instruction resource library, and the code example is shown in fig. 7. The logic in this decision class is roughly: firstly, according to the parameters: the input value of "sender", such as "Pebax" in this example, goes to the database tables of "personal client" and "enterprise client", respectively, to make inquiries, if in the client resource table, there are matched data records, then the waybill will be created as sender based on the client data of this type, if there are no matched data records, then the "sender" input value based on the parameter, as zero-time client, and as sender, to implement creating waybill, and in the course of waybill creation, there is a need to create zero-time client at the same time.

Suppose that in the actual system operation, when the user inputs the information of the shipper as "Pectare Cola", there are several scenarios:

scene 1: after logical judgment, the mapping decision class finds that enterprise-level client resources named as 'Baishi Cola' exist in a database table of the background 'enterprise client', the parameter resources need to be redirected from a general delivery party parameter 'delivery party (sender)' to a parameter resource 'sender corporation' of a standard enterprise delivery party, and the value of the parameter is set to 'Baishi Cola', so that the existing enterprise resource 'Baishi Cola' can be used as the delivery party, and freight note creation is implemented.

Scene 2: assuming that in actual operation, a background "personal client" database table is found to have a resource named "pepercola", a parameter resource "senderIndividual" needs to be defined to the personal client, and the value of the parameter is set to be "pepercola".

Scene 3: assuming that in actual operation, resources are matched in both the "individual customer" and "enterprise customer" resources, then redirection to the two parameter resources "enterprise customer shipper sendercorporation" and "individual customer shipper senderinduvidual" is required and two secondary response instructions are generated. Finally, fed back to the user's interface, two executable instructions appear: one is based on the enterprise client 'Baishi Cola' existing in the system as the delivery party to establish the waybill, and the other is based on the personal client 'Baishi Cola' existing in the system as the delivery party to establish the waybill. The client can select according to own will.

Scene 4: assuming that no resource can be found in the resources of the "personal client" and the "enterprise client" simultaneously in the time operation, and matching is realized with the input value of the "hectareca", according to the logic of the "mapping decision class", the following two parameters, namely, client information of the "hectareca", which is the enterprise goods sender at the time of creating the waybill, and client information of the "hectareca", which is the personal goods sender at the time of creating the waybill, are required to be redirected to the two parameters, namely, client information of the "hectareca", which is the enterprise goods sender at the time of creating the waybill, or client information of the "hectareca", which is the personal goods sender at the time of creating the waybill, and the user can select the sentence required to be executed according to the own intention.

It can be seen from this that: defining the mapping decision class on the parameter resource provides the opportunity for interacting the user input with the background logic and background resource. The analysis engine can further judge and simulate the intention of the user according to the input of the user and the actual situation of the system resources, and feeds back the intention to the real executable scheme for the client to select, so that the execution of the client can be more accurate. On the other hand, the definition has certain ambiguity and ambiguity like a delivery side and a receiving side, and the mapping decision class on the resource node is not directly executed, so that the support for inputting the ambiguity and the ambiguity parameters by the user is realized.

The process of instruction resource combination:

according to the description of the foregoing analysis process, after the instruction analysis, the resource obtained by the system through the instruction analysis exists in the data of the composite structure, such as the "secondary response instruction", and cannot be directly fed back to the user or directly executed. A combined process is required.

On one hand, the current system framework uniformly executes the user's instructions in the manner of AJAX format instructions, and the final execution results of the instructions are uniformly relocated to the pages that the client needs to execute. Such as the example of "create waybill" given in the previous section, the final result of the instructions fed back to the user is to relocate to the page where the waybill was created and where the parameters are set according to the user's input, such as the example where the shipper of the waybill is set to "Pebax". The instruction execution format is a unified AJAX format, and the instruction can be directly executed, and two basic factors are needed: the address of the execution instruction and the parameters of the instruction.

The composition of the execution address: the parameters can be generated by combining link resources (linkURL) contained in the target and command resources in the secondary response instruction, and the parameters are generated by the final parameter resources after the instruction is parsed. In the logistics management system based on the JAVA platform, the command in the AJAX format executes the link generation format as follows: html < engineering link address >/< destination link address >/< command link address >. For example, in the previous example of creating a waybill, the entire address may be: [ localhost: 8080/projectName/bookmark/create. The arguments of the AJAX instruction may be formed directly from the key-value pairs consisting of the names and values of all the command arguments that may ultimately be executed.

On the other hand: the system also needs to describe the instruction according to an accurate and readable language and feed the instruction back to the client, and the instruction is provided for the client to make further judgment and selection: also in the instruction resource pool, each node resource: the command resource, the target resource and the parameter resource all have corresponding fields, namely, label (display text), for marking the description information of the resource, and the instruction analysis engine can combine the display texts of the resources according to a certain rule and display the combined display texts on the user interface.

For example, in the foregoing example of creating an waybill, the display text corresponding to the target resource is "waybill", the display text corresponding to the command resource is "create", and when the shipper "baccara" is mapped to two parameters, namely "enterprise customer shipper" sendee "and" personal customer shipper sendee "at the same time, their corresponding display texts are [ shipper is enterprise customer"% S "] and [ shipper is personal customer"% S "], respectively, where"% S "is a placeholder, and the runtime is populated with parameter values, two executable commands are simultaneously displayed on the final user interface, and the texts are:

1. and (4) creating a waybill, wherein a shipper is an enterprise client 'PebaC Cola'.

2. A manifest is created and the shipper is the individual customer "pepsi cola".

If the shipper "PeterCola" does not find a customer resource match in the system, and the next two parameters "establish the Enterprise customer shipper tmpender corporation" and "establish the Individual customer shipper tmpender corporation" are considered, their corresponding display texts are respectively [ shipper when creating zero ]: enterprise customer "% S" ] and [ shipper when creating time ": personal client "% S". The text displayed by the user interface becomes:

1. creating a waybill, and creating a zero-hour shipper: the enterprise client is 'Baishi Cola'.

2. Creating a waybill, and creating a zero-hour shipper: the individual customer, "Peterse Cola".

Description of the drawings:

fig. 1 shows the whole process of the intelligent semantic analysis instruction robot.

FIG. 2 illustrates a data model structure of an instruction repository.

FIG. 3 illustrates, by way of example, a simple flow of matching a target resource and a command resource during semantic analysis.

FIG. 4 shows a data model structure of a compound parameter instruction resource.

Fig. 5 shows a data model structure of a composite parameter structure, taking "cargo weight" as an example.

Fig. 6 shows the implementation code of the "secondary matching agent class" for the weight unit of goods "kg" in the semantic analysis process.

Fig. 7 shows implementation codes of "control decision classes" for controlling the fuzziness parameter of waybill "shipper" in the instruction analysis process.

FIG. 8 shows the control flow of the mapping of instruction response resources during instruction analysis.

Fig. 9 shows an interface for setting instruction resources of the instruction resource library in the logistics management system.

Fig. 10 shows an interface for setting a "primary response resource" of the instruction resource library in the logistics management system.

Fig. 11 shows an interface for operating the intelligent instruction robot in the logistics management system.

Fig. 12 shows the result of operating the intelligent instruction robot and clicking the executable command in the logistics management system.

The specific implementation mode is as follows:

the specific implementation process of the system comprises the following steps: how a developer extends an instruction resource library under the framework of an instruction analysis engine; the method comprises the steps of expanding primary response instruction resources of an instruction resource library to realize semantic analysis of commands input by a user in natural language, and expanding secondary response instruction resources of the instruction resource library to realize mapping and combination of instructions and finally feed back the instructions to a client to execute only AJAX commands and explanatory texts. Finally, how an operator operates the intelligent instruction robot and the operation effect in the system is introduced on the application level.

How to expand the extended command library:

when the system has new functions and the intelligent command robot is required to implement and control the new functions, developers need to register the new functions as system command resources in a command database under a command analysis engine.

The registration process is as follows: an operator firstly logs in a system instruction resource management interface and enters a system instruction target resource searching page through a navigation bar (system management) to (instruction resource management). The user can enumerate and search all the instruction target resources of the system. The data structure of the instruction resource library is managed according to a tree structure of "target" - "command" - "parameter", as shown in fig. 2, the target resource is a root directory, the parameter resource is subordinate to the command resource, and the command resource is subordinate to the target resource. Therefore, to operate on the parameter resource or the command resource, the operation must be performed from the related upper level resource.

Editing operation on the target resource: if the target resource needs to be newly added, "newly added" is clicked in the target resource searching page, the editing interface of the instruction resource is entered, and if the existing target resource needs to be edited, the target resource needing to be edited is selected from the list and the editing interface of the resource is entered. In the target resource editing interface, the user needs to specify the number ID, link, and display text of the target resource, which is needed in the process of combining the resources.

For example, the target resource of the waybill of the system, the resource number ID is "BookingNote", the link is "BookingNote", and the display text is "waybill". Meanwhile, the editing interface of the target resource displays all command resources belonging to the target resource, such as basic instruction resources of 'new construction', 'editing', 'searching', 'viewing', and the like, below a resource editing page of 'waybill' at the same time.

Editing operation on command resources: if the user needs to add a new command resource or edit the existing command resource, the user needs to firstly confirm the instruction target resource of the previous stage and click the 'new' button or the command resource on the command list below the target resource. Similarly, if the parameter under a certain command needs to be set, the editing page of the command resource needs to be entered first, and then the parameter resource is added or edited in the following parameter setting part.

If the parameter has a transport symbol or unit, a newly added 'parameter unit' button and a 'parameter operator' button are clicked to realize the newly added unit and parameter operator resources in the editing interface.

If the parameter can be directly used as the parameter submitted when executing the command, only the name and the display text of the parameter are required to be set, and the mapping decision class is not required to be set, if the parameter has logic of further mapping, when the parameter is required to be further mapped into the parameter of the next level of direct operation in the analysis process, the mapping rule class is required to be defined, the mapping rule is realized, and the full name of the mapping decision class is set to the field of the parameter resource: mapping decision classes above.

And (3) mapping the code implementation flow of the rule class: the standard interface of 'instruction mapping decision class' needs to be realized: [ ILanuageComParaMappingDesicion ], and implements the core method: [ executemaping ] is provided. An example of mapping decision classes is shown in FIG. 7. The example implements code for mapping decision class of parameter "sender", and the core method of this class is: matching the consignor parameter value contained in the primary response instruction with the enterprise client and personal client resources in the system, dividing the primary response instruction into 4 scenes according to different conditions, and mapping the current parameters to the parameters of the next level in sequence: and the personal shipper, the enterprise shipper, the personal shipper and the enterprise shipper, the creation personal shipper and the creation enterprise shipper and the like are different in the next-level parameters so as to combine the parameters, the commands and other resources to form finally executable parameters.

How to expand the primary response resource portion of the instruction resource pool:

the data structure of the instruction resource library is shown in fig. 2, the core part, i.e. the secondary response resource part, is managed according to a tree structure of "target" - "command" - "parameter", and each level of "resource node" includes a "primary response resource" node for semantic analysis and matching by a semantic analysis engine. In the current system framework, a "resource node" comprises a "primary response resource" node, and is considered to be extended to a plurality of "primary response resource" nodes.

Editing operation on the node of the 'primary response resource': and no additional operation is needed, after entering the editing interface of each level of instruction target resource, the node part of the 'primary response resource' can automatically appear, and after direct selection, the editing button is clicked to enter the editing interface.

Wherein, the field "match content": the keywords are used for editing the keywords matched by the semantic analysis engine, if a plurality of keywords exist, the keywords are separated by commas, the operation interface is shown in fig. 9, and the weight unit of goods "kg" contains 4 matched keywords: when the four keywords are included in the input sentence of the user, the four keywords are considered to be successfully matched with the node of the weight unit of goods of kilogram.

The field "matching degree" is used to indicate the matching percentage when the keywords match, and the default matching degree is 80%: that is, when there are 10 matched keywords, the matching can be considered as successful by being able to contact 8 matched words.

If it is necessary to define the "secondary matching logic" such as matching from the cargo weight unit "kg" to the cargo weight and the cargo name through the secondary matching, the developer is required to customize the "secondary matching agent class nextmappingproxycalass" and set it on the "primary response resource" node, as shown in fig. 9. When the code implements the "second matching agent class nextmappingproxycalass", it first needs to inherit the standard interface [ ILanuageNextCommandMappingProxy ] and implement the logic of second matching inside the core method maptoconextcommand.

The user operates the semantic robot in the system:

the system is logged in first, and a user work center page can be accessed through a navigation bar (a user center). By clicking a button, the semantic analysis robot is started, and a dialog box interface pops up, and fig. 11 shows that the intelligent robot of the logistics management system starts a page. The user can enter instructions of the language type in the page and parameters can be added. As shown, when the user enters: "create waybill, shipper: when the goods are 500kg cola, after the system analyzes, two executable schemes are fed back to a user according to the actual data situation of the system, namely creating a waybill, wherein a delivery party is an enterprise client (Baishi Cola Chengdu company), the goods name is [ Cola ], the weight is [ 500 ] kg "and the creating waybill, and the delivery party is an enterprise client [ Baishi Cola Chengdu Mipu factory ], the goods name is [ Cola ], and the weight is [ 500 ] kg. When the user selects the first execution scheme and clicks the execution link, the page of the system jumps to the waybill creation page, and the information of the enterprise shipper and the information of the goods, including the name of the goods, the weight of the goods and the like, are automatically filled into the waybill, so that the time of the operator is saved, and the operator unfamiliar with the system can be more quickly familiar with the functions of the system.

Claims (7)

1. The invention constructs the intelligent language operation interface of the logistics management software by inventing the semantic analysis engine and the instruction analysis engine based on the physical management software platform, and a user can directly operate the management software by inputting the instruction of the natural language type in the general interface.

2. Based on the intelligent language operation interface described in claim one, the instruction analysis engine can deduce all possible execution intentions of the user according to the language type input of the user after analysis, and generate one or more executable schemes by combining the actual resource situation of the system, including clicking the link for executing the AJAX format command, with related parameters, and the description information corresponding to each executable scheme, and feeding back the link on the general operation interface for the user to select and execute.

3. Based on the instruction analysis engine described in claim one, all the operable instruction resources of the system are stored and managed in the instruction resource library of the logistics management software according to the tree structure data model of "instruction target- > operation command- > instruction parameter", and the node of each tree structure contains information for semantic analysis matching and information for executable instruction construction.

4. Based on the instruction resource model described in the claim three, any level of instruction resource can support ambiguous and ambiguous instruction resources input by a user, including ambiguous operation targets, operation commands and command parameters, and developers can realize mapping and relocation of the ambiguous instruction resources to one or more precise instruction resources by customizing the mapping decision classes on the resource model nodes.

5. Based on the instruction resource model described in claim three, the "command parameters" resource node is followed by the lower level resource nodes: and the 'parameter operator' and the 'parameter unit' resource node are used for realizing the matching of the composite parameter resource.

6. Based on the instruction resource model described in claim three: all the data nodes of each level of the tree structure of 'instruction target- > operation command- > instruction parameter' contain the address information and the description information required for constructing the AJAX format instruction, and the instruction analysis engine combines the address information and the description information of the nodes into the final execution address and the final description information of the AJAX command required in the executable scheme according to a certain mode.

7. Based on the instruction resource model described in claim three, each data model node contains basic information required for semantic analysis and keyword matching: the matching degree is set to control the matching percentage of the keywords, one or more keywords are set according to the matching content, and the complex semantic matching relation is realized through the secondary matching agent class.