Nothing Special   »   [go: up one dir, main page]

CN110781685A - Method for automatically marking correctness of semantic analysis result based on user feedback - Google Patents

Method for automatically marking correctness of semantic analysis result based on user feedback Download PDF

Info

Publication number
CN110781685A
CN110781685A CN201910998224.9A CN201910998224A CN110781685A CN 110781685 A CN110781685 A CN 110781685A CN 201910998224 A CN201910998224 A CN 201910998224A CN 110781685 A CN110781685 A CN 110781685A
Authority
CN
China
Prior art keywords
data
semantic analysis
cache
current
marked
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN201910998224.9A
Other languages
Chinese (zh)
Other versions
CN110781685B (en
Inventor
刘楚雄
夏承建
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Sichuan Changhong Electric Co Ltd
Original Assignee
Sichuan Changhong Electric Co Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Sichuan Changhong Electric Co Ltd filed Critical Sichuan Changhong Electric Co Ltd
Priority to CN201910998224.9A priority Critical patent/CN110781685B/en
Publication of CN110781685A publication Critical patent/CN110781685A/en
Application granted granted Critical
Publication of CN110781685B publication Critical patent/CN110781685B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Landscapes

  • Machine Translation (AREA)

Abstract

The invention discloses a method for automatically marking the correctness of semantic analysis results based on user feedback, which effectively utilizes mass data generated by semantic analysis, automatically marks the correctness of the semantic analysis results based on different feedback behaviors made by a user in response to a system, saves a large amount of manual marking cost and improves the marking efficiency through big data analysis and weight processing.

Description

Method for automatically marking correctness of semantic analysis result based on user feedback
Technical Field
The invention relates to the technical field of semantic annotation, in particular to a method for automatically annotating the correctness of a semantic analysis result based on user feedback.
Background
Semantic analysis is an important direction in the field of computer natural language processing, aiming at mapping natural language used by human beings into a complete, computer-executable formal meaning representation. The method is a core research field of natural language processing, and is also a key technology for realizing various intelligent systems, such as a human-computer interaction system, voice control of intelligent equipment, robot action control and the like. The main flow of semantic analysis work is to classify the text data into domains, then to adopt the corresponding models of each domain to identify intentions, identify entities, analyze syntax, disambiguate semantics and so on according to the domain classification result, and finally to synthesize the analysis result to form the semantic analysis result.
Most of the semantic analysis systems actually falling to the ground are continuous iterative items, and need to be continuously optimized and adjusted. The optimization work mainly aims at analyzing data generated by a user in the using process, and the analysis efficiency can be improved by carrying out model or process optimization on the analyzed correct data. And analyzing the data with the analysis error can find the new requirements of the system BUG and the user in time, thereby improving the accuracy and the coverage of the system analysis. The judgment of the correctness of the analysis data comes from manual labeling, and the method consumes a large amount of manpower and is not beneficial to development and iteration. At present, the research of the industry on semantic automatic labeling focuses on the level of part of speech labeling and syntax labeling, and an efficient and accurate method for carrying out correct and wrong labeling on analyzed data is lacked.
Disclosure of Invention
Most of the current floor-type semantic analysis systems are terminal-oriented, and after a user initiates a semantic analysis request once, a semantic analysis result is finally fed back to the user through the terminal. The correct response, the incorrect response, the delayed response or the non-response generated by the system can lead the user to have continuous semantic request behavior to feed back to the system through visual perception, and the correctness of each semantic analysis result of each user can be marked by analyzing the semantic analysis result difference, the time difference, the field difference and the emotion difference among the feedback of the users in the same round of conversation (the same round of conversation in the patent refers to the conversation with the same field in a certain time). Because each user has different feedback habits, the correctness of the semantic analysis result can be finally marked by combining the mass data analysis and the weight calculation. The method and the device utilize data generated in the actual using process of the user, and realize automatic correct and wrong labeling of semantic analysis results according to feedback of the user to different responses, save a large amount of manual labeling cost, and improve the labeling efficiency. And simultaneously, a large amount of marked data can be generated for scientific research and production.
The invention realizes the purpose through the following technical scheme:
the method for automatically marking the correctness of the semantic analysis result based on the user feedback comprises the following steps:
step 1, determining the tolerable error response times of a user in the same round of conversation;
step 2, collecting data generated by a semantic analysis system, summarizing all field intentions related to the data, and determining that two adjacent semantic requests are the longest time interval of the same round of conversation and field groups which can be regarded as the same round of conversation according to different field intentions;
step 3, aggregating the data according to different users, and arranging the data of the same user in an ascending order according to the request time;
step 4, traversing the data which are sequenced by each user, sequentially taking out each piece of data, and comparing and analyzing the data with the latest piece of data in the cache region;
(1) judging whether the data in the current cache area is empty, if so, caching the current data and acquiring the next piece of data;
(2) taking out the data closest to the current data in the cache region, judging the field difference and the time difference of the two data, determining whether the two data are in the same round of conversation, if so, jumping to (3), otherwise, jumping to (4);
(3) whether the current request is clear evaluation on the semantic analysis result is analyzed, if the current request is positive evaluation, skipping (5) is carried out, if the current request is negative evaluation, skipping (6) is carried out, and if the current request is not clear evaluation, the current data is cached;
(4) and at the moment, the current data and the data in the cache are dialogues in different turns, firstly, whether the semantic analysis results of the data in the cache are consistent or not is judged, and if so, skipping (6). Otherwise, judging the quantity of the data in the cache, if the quantity is larger than the tolerable error response times of the user in the same conversation, marking all the data in the cache as errors, emptying the cache, then putting the current data in the cache, and otherwise, skipping (5);
(5) at this moment, the last piece of data is correct, namely one piece of data adjacent to the current data in the cache is marked as correct, and the rest pieces of data are marked as errors, emptying the cache and storing the current data;
(6) and at this moment, all the previous data are wrong, namely all the data in the cache are marked as wrong, and the cache is emptied and the current data is stored.
Step 5, defining a mark identified as the same semantic analysis data, summarizing all marked single data according to the mark, and counting the marked correct frequency t and the error frequency f of the same semantic analysis data;
step 6, if the accuracy of the semantic analysis system exists, the semantic analysis data can be finally labeled by adopting a strategy;
if not, the correct and wrong frequencies can be directly compared, and the larger frequency is selected as the label.
Further, in the step 6, the strategy adopted for finally labeling the semantic analysis data is as follows:
tp is more than or equal to f (1-p) is marked as correct;
tp < f (1-p) is marked as error;
wherein p is the accuracy of the semantic analysis system.
The invention has the beneficial effects that:
the method for automatically marking the correctness of the semantic analysis result based on the user feedback effectively utilizes the mass data generated by the semantic analysis, automatically marks the correctness of the semantic analysis result based on different feedback behaviors made by the user in response to the system, and then saves a large amount of manual marking cost and improves the marking efficiency through the big data analysis and the weight processing.
Drawings
In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the embodiments or the drawings needed to be practical in the prior art description, and obviously, the drawings in the following description are only some embodiments of the embodiments, and for those skilled in the art, other drawings can be obtained according to these drawings without creative efforts.
FIG. 1 is a flow chart of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the technical solutions of the present invention will be described in detail below. It is to be understood that the described embodiments are merely exemplary of the invention, and not restrictive of the full scope of the invention. All other embodiments, which can be derived by a person skilled in the art from the examples given herein without any inventive step, are within the scope of the present invention.
In any embodiment, as shown in fig. 1, the common smart TV is used as a terminal to explain the present embodiment, fifty fields and one hundred intentions are involved in the actual use process of the floor semantic analysis system, and the present solution only selects some common fields, VIDEO, MUSIC, CONTROL, TV, CHAT, and intent QUERY, pay, record, email, SET, CHAT. When the data can not be classified into video, music and television fields, the data is uniformly divided into chatting.
Log data for a day is intercepted as follows, with some fields omitted or simplified:
{ id: "A", mac: 18-99-f5-f4-ba-33 ", time: "2019-09-0820: 40: 33 ", text: "special soldier is frontal," domain: "CHAT", intent: "CHAT", semantic: {}}
{ id: "B", mac: 18-99-f5-f4-ba-33 ", time: "2019-09-0820: 40: 40 ", text: "want to see that i is a special soldier", domain: "VIDEO", intent: "QUERY", semantic: { name: "I am a special soldier" }
{ id: "C", mac: 18-99-f5-f4-ba-33 ", time: "2019-09-0820: 40: 48 ", text: "first collection troublesome to play", domain: "CONTROL", intent: "EPISODE", semantic: { action: "index", index: "1"}}
{ id: "D", mac: 18-99-f5-f4-ba-33 ", time: "2019-09-0820: 40: 54 ", text: "smart o", domain: "CHAT", intent: "CHAT", semantic: {}}
{ id: "E", mac: 18-99-f5-f4-ba-33 ", time: "2019-09-0823: 30: 12 ", text: "turn off tv", domain: "TV", intent: "SET", semantic: { operatands: "OBJ _ SETTING _ CLOSE" }
{ id: "F", mac: "00-6 c-ff-ee-75-ad", time: "2019-09-0820: 50: 05 ", text: "you are my answers," domain: "MUSCI", intent: "PLAY", semantic: { song: "you are my peace" }
{ id: "G", mac: "00-6 c-ff-ee-75-ad", time: "2019-09-0820: 50: 25 ", text: "i want to see you are my answers," domain: "VIDEO", intent: "QUERY", semantic: { name: "you are my eyes" }
{ id: "H", mac: "00-6 c-ff-ee-75-ad", time: "2019-09-0820: 50: 47 ", text: "Play your answer TV show", domain: "VIDEO", intent: "QUERY", semantic: { name: "you are My eyes", category: "TV play" }
{ id: "I", mac: "00-6 c-ff-ee-75-ad", time: "2019-09-0820: 50: 57 ", text: "you are my answer tv drama", domain: "VIDEO", intent: "QUERY", semantic: { name: "you are My eyes", category: "TV play" }
{ id: "J", mac: "00-6 c-ff-ee-75-ad", time: "2019-09-0820: 51: 11 ", text: "turn off tv", domain: "TV", intent: "SET", semantic: { operatands: "OBJ _ SETTING _ CLOSE" }
1. Determining the tolerance times of the user to the error response according to the terminal:
here defined as 3.
2. The maximum time interval for each domain intent to be determined for the same round of conversation is as follows (unit: seconds):
Figure BDA0002239231320000051
VIDEO and MUSIC relate to the field that network access is required to request resources, and users also need to spend time on analyzing and feeding back the resources, so the time setting is longer than CONTROL, CHAT, TV and the like for timely response. Wherein PLAY is also intended to require buffering of data PLAY, longer time should be set.
Both the VIDEO and MUSIC fields are related to entity data, so allowing both to occur in the same wheel-to-wheel conversation. The CHAT domain data is data that is not recognized by the domain classification and may therefore occur in any one round of the session. Based on the above analysis, the domains identifiable as the same round of conversation are grouped as [ VIDEO, MUSIC, CHAT ], [ CONTROL, CHAT ], [ TV, CHAT ], [ CHAT ].
3. The map-reduce is used for converging data according to the mac fields (determined as uniform users according to the mac), and the data are arranged in an ascending order according to the time fields, so that the following results are obtained:
{mac:”18-99-f5-f4-ba-33”,logsId:[“A”,”B”,”C”,”D”]}
{mac:”00-6c-ff-ee-75-ad”,logsId:[“E”,”F”,”G”,”H”,”I”]}
4. analyzing all converged ordered data:
the data for mac addresses 18-99-f5-f4-ba-33 were first analyzed:
and (4) acquiring a first piece of data A, and directly storing the first piece of data A into a cache according to the step (1).
And (3) acquiring the next piece of data B, classifying the current data field as VIDEO according to the step 4 (2), classifying the latest one in the cache as CHAT, wherein the VIDEO and the CHAT are in the field group of the same round of conversation, and the time interval of the two pieces of data is 7 seconds and less than 10 seconds, so that the two pieces of data belong to the same round of conversation, and entering the step 4 (3), wherein the current data is not evaluated and the current data is cached.
Acquiring the next piece of data C, classifying the current data field as CONTROL according to the step 4, (2), and classifying the current data field as VIDEO in the cache, so that the current data field does not belong to the same round of conversation, wherein A, B pieces of data exist in the cache region, and according to the step 4, (4), the semantic analysis result of A is sematic: { }, the semantic analysis result of B is semantic: { name: and (4) performing step (5) when the two are inconsistent, marking A as an error and B as a correct, clearing the buffer area and caching the data C.
And acquiring the next piece of data D, wherein the data D and the data in the cache belong to the same round of conversation according to the step 4, (2), the data D is positively evaluated according to the step 4, (3), the step 4), (5) is carried out, and the data D is marked as correct, and the data D is buffered in a buffer area.
And (3) acquiring the next piece of data E, wherein according to the step 4, (2), if the E and the data in the cache do not belong to the same round of conversation, according to the step 4, (4), marking D as correct, clearing the cache region and caching the data E.
Finally, the time interval between data E and the next day zero also exceeds the time of the same session, so according to step 4 (4), E is marked as correct.
Continue to analyze the data with mac address 00-6 c-ff-ee-75-ad:
and (4) acquiring a first piece of data F, and directly storing the first piece of data F into a cache according to the step (4).
And (3) acquiring the next piece of data G, wherein the current data and the cache data belong to the same round of conversation according to the step 4 (2), and entering the step 4 (3), wherein the current data is not evaluated, and the data G is cached.
And acquiring the next piece of data H, and performing the same analysis process as the data G.
And acquiring the next piece of data I, and performing the same analysis process as the data G.
And (3) acquiring the next piece of data J, wherein the current data and the cache data do not belong to the same round of conversation according to the step 4.(2), and entering the step 4.(4), wherein E, F, H, G are marked as errors because the number of the data in the cache is 4 and is greater than the tolerance 3.
Finally, the interval between data I and the next day zero exceeds the time of the same session, so I is labeled as correct according to step 4 (4).
5. Converging the marked data according to the contents of text, domain, intent and semantic fields, summarizing the marked data into correct and wrong frequencies, and analyzing the data as follows (true: correct frequency, false: wrong frequency):
{ id: "a", text: "special soldier is frontal," domain: "CHAT", intent: "CHAT", semantic: { }, true: 0, false: 1}
{ id: "b", text: "want to see that i is a special soldier", domain: "VIDEO", intent: "QUERY", semantic: { name: "I am a special soldier" }, true: 1, false: 0}
{ id: "c", text: "first collection troublesome to play", domain: "CONTROL", intent: "EPISODE", semantic: { action: "index", index: "1" }, true: 1, false: 0}
{ id: "d", text: "smart o", domain: "CHAT", intent: "CHAT", semantic: { }, true: 1, false: 0}
{ id: "e", text: "turn off tv", domain: "TV", intent: "SET", semantic: { operatands: "OBJ _ SETTING _ CLOSE" }, true: 2, false: 0}
{ id: "f', text: "you are my answers," domain: "MUSCI", intent: "PLAY", semantic: { song: "you are my peace" }, true: 0, false: 1}
{ id: "g", text: "i want to see you are my answers," domain: "VIDEO", intent: "QUERY", semantic: { name: "you are my eyes" }, true: 0, false: 1}
{ id: "h", text: "Play your answer TV show", domain: "VIDEO", intent: "QUERY", semantic: { name: "you are My eyes", category: "tv drama" }, true: 0, false: 1}
{ id: "i", text: "you are my answer tv drama", domain: "VIDEO", intent: "QUERY", semantic: { name: "you are My eyes", category: "tv drama" }, true: 0, false: 1}
6. The experience accuracy of the semantic analysis system carried by the smart television is about 0.65, and the final label of each piece of analytic data can be calculated by substituting a formula as follows:
data a, with 0 x 0.7 < 1 x (1-0.7), marked as error;
data b, c, d, with 1 x 0.7 > 0 x (1-0.7), are labeled as correct;
data e, with 2 x 0.7 > 0 x (1-0.7), labeled correct;
data f, g, h, i, with 0 x 0.7 < 1 x (1-0.7), marked as error;
the above description is only for the specific embodiments of the present invention, but the scope of the present invention is not limited thereto, and any person skilled in the art can easily conceive of the changes or substitutions within the technical scope of the present invention, and all the changes or substitutions should be covered within the scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the appended claims. It should be noted that the various technical features described in the above embodiments can be combined in any suitable manner without contradiction, and the invention is not described in any way for the possible combinations in order to avoid unnecessary repetition. In addition, any combination of the various embodiments of the present invention is also possible, and the same should be considered as the disclosure of the present invention as long as it does not depart from the spirit of the present invention.

Claims (3)

1. The method for automatically marking the correctness of the semantic analysis result based on the user feedback is characterized by comprising the following steps of:
step 1, determining the tolerable error response times of a user in the same round of conversation;
step 2, collecting data generated by a semantic analysis system, summarizing all field intentions related to the data, and determining that two adjacent semantic requests are the longest time interval of the same round of conversation and field groups which can be regarded as the same round of conversation according to different field intentions;
step 3, aggregating the data according to different users, and arranging the data of the same user in an ascending order according to the request time;
step 4, traversing the data which are sequenced by each user, sequentially taking out each piece of data, and comparing and analyzing the data with the latest piece of data in the cache region;
step 5, defining a mark identified as the same semantic analysis data, summarizing all marked single data according to the mark, and counting the marked correct frequency t and the error frequency f of the same semantic analysis data;
step 6, if the accuracy of the semantic analysis system exists, the semantic analysis data can be finally labeled by adopting a strategy;
if not, the correct and wrong frequencies can be directly compared, and the larger frequency is selected as the label.
2. The method for automatically labeling correctness of semantic analysis results based on user feedback according to claim 1, wherein in the step 4, the step of comparing and analyzing the data to be fetched and the latest data in the cache area comprises:
(1) judging whether the data in the current cache area is empty, if so, caching the current data and acquiring the next piece of data;
(2) taking out the data closest to the current data in the cache region, judging the field difference and the time difference of the two data, determining whether the two data are in the same round of conversation, if so, jumping to (3), otherwise, jumping to (4);
(3) whether the current request is clear evaluation on the semantic analysis result is analyzed, if the current request is positive evaluation, skipping (5) is carried out, if the current request is negative evaluation, skipping (6) is carried out, and if the current request is not clear evaluation, the current data is cached;
(4) and at the moment, the current data and the data in the cache are dialogues in different turns, firstly, whether the semantic analysis results of the data in the cache are consistent or not is judged, and if so, skipping (6). Otherwise, judging the quantity of the data in the cache, if the quantity is larger than the tolerable error response times of the user in the same conversation, marking all the data in the cache as errors, emptying the cache, then putting the current data, and otherwise, skipping (5);
(5) at this moment, the last piece of data is correct, namely one piece of data adjacent to the current data in the cache is marked as correct, and the rest pieces of data are marked as errors, emptying the cache and storing the current data;
(6) and at this moment, all the previous data are wrong, namely all the data in the cache are marked as wrong, and the cache is emptied and the current data is stored.
3. The method for automatically labeling the correctness of the semantic analysis result based on the user feedback as claimed in claim 1, wherein in the step 6, the strategy for performing the final labeling on the semantic analysis data is as follows:
tp is more than or equal to f (1-p) is marked as correct;
tp < f (1-p) is marked as error;
wherein p is the accuracy of the semantic analysis system.
CN201910998224.9A 2019-10-18 2019-10-18 Method for automatically marking correctness of semantic analysis result based on user feedback Active CN110781685B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201910998224.9A CN110781685B (en) 2019-10-18 2019-10-18 Method for automatically marking correctness of semantic analysis result based on user feedback

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201910998224.9A CN110781685B (en) 2019-10-18 2019-10-18 Method for automatically marking correctness of semantic analysis result based on user feedback

Publications (2)

Publication Number Publication Date
CN110781685A true CN110781685A (en) 2020-02-11
CN110781685B CN110781685B (en) 2022-08-19

Family

ID=69386131

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201910998224.9A Active CN110781685B (en) 2019-10-18 2019-10-18 Method for automatically marking correctness of semantic analysis result based on user feedback

Country Status (1)

Country Link
CN (1) CN110781685B (en)

Citations (16)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN1294812A (en) * 1996-11-18 2001-05-09 Mci全球通迅公司 Communication system architecture
CN1853180A (en) * 2003-02-14 2006-10-25 尼维纳公司 System and method for semantic knowledge retrieval, management, capture, sharing, discovery, delivery and presentation
CN101446942A (en) * 2008-12-10 2009-06-03 苏州大学 Semantic character labeling method of natural language sentence
CN101706792A (en) * 2009-11-27 2010-05-12 河南大学 Chinese query clause oriented three-level inquired target analysis method
CN101763574A (en) * 2009-06-03 2010-06-30 中国科学院自动化研究所 Historic building conservation technical information management system and method based on domain knowledge
CN101908042A (en) * 2010-08-09 2010-12-08 中国科学院自动化研究所 Tagging method of bilingual combination semantic role
CN104462041A (en) * 2014-11-28 2015-03-25 上海埃帕信息科技有限公司 Method for completely detecting hot event from beginning to end
CN105574038A (en) * 2014-10-16 2016-05-11 阿里巴巴集团控股有限公司 Text content recognition rate test method and device based on anti-recognition rendering
US20160253309A1 (en) * 2015-02-26 2016-09-01 Sony Corporation Apparatus and method for resolving zero anaphora in chinese language and model training method
CN107480196A (en) * 2017-07-14 2017-12-15 中国科学院自动化研究所 A kind of multi-modal lexical representation method based on dynamic fusion mechanism
CN108417205A (en) * 2018-01-19 2018-08-17 苏州思必驰信息科技有限公司 Semantic understanding training method and system
EP3373156A1 (en) * 2015-11-04 2018-09-12 Alibaba Group Holding Limited Method and device for dividing webpage tabs into groups
CN109036388A (en) * 2018-07-25 2018-12-18 李智彤 A kind of intelligent sound exchange method based on conversational device
CN109800296A (en) * 2019-01-21 2019-05-24 四川长虹电器股份有限公司 A kind of meaning of one's words fuzzy recognition method based on user's true intention
CN109817329A (en) * 2019-01-21 2019-05-28 暗物智能科技(广州)有限公司 A kind of medical treatment interrogation conversational system and the intensified learning method applied to the system
CN110209791A (en) * 2019-06-12 2019-09-06 百融云创科技股份有限公司 It is a kind of to take turns dialogue intelligent speech interactive system and device more

Patent Citations (16)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN1294812A (en) * 1996-11-18 2001-05-09 Mci全球通迅公司 Communication system architecture
CN1853180A (en) * 2003-02-14 2006-10-25 尼维纳公司 System and method for semantic knowledge retrieval, management, capture, sharing, discovery, delivery and presentation
CN101446942A (en) * 2008-12-10 2009-06-03 苏州大学 Semantic character labeling method of natural language sentence
CN101763574A (en) * 2009-06-03 2010-06-30 中国科学院自动化研究所 Historic building conservation technical information management system and method based on domain knowledge
CN101706792A (en) * 2009-11-27 2010-05-12 河南大学 Chinese query clause oriented three-level inquired target analysis method
CN101908042A (en) * 2010-08-09 2010-12-08 中国科学院自动化研究所 Tagging method of bilingual combination semantic role
CN105574038A (en) * 2014-10-16 2016-05-11 阿里巴巴集团控股有限公司 Text content recognition rate test method and device based on anti-recognition rendering
CN104462041A (en) * 2014-11-28 2015-03-25 上海埃帕信息科技有限公司 Method for completely detecting hot event from beginning to end
US20160253309A1 (en) * 2015-02-26 2016-09-01 Sony Corporation Apparatus and method for resolving zero anaphora in chinese language and model training method
EP3373156A1 (en) * 2015-11-04 2018-09-12 Alibaba Group Holding Limited Method and device for dividing webpage tabs into groups
CN107480196A (en) * 2017-07-14 2017-12-15 中国科学院自动化研究所 A kind of multi-modal lexical representation method based on dynamic fusion mechanism
CN108417205A (en) * 2018-01-19 2018-08-17 苏州思必驰信息科技有限公司 Semantic understanding training method and system
CN109036388A (en) * 2018-07-25 2018-12-18 李智彤 A kind of intelligent sound exchange method based on conversational device
CN109800296A (en) * 2019-01-21 2019-05-24 四川长虹电器股份有限公司 A kind of meaning of one's words fuzzy recognition method based on user's true intention
CN109817329A (en) * 2019-01-21 2019-05-28 暗物智能科技(广州)有限公司 A kind of medical treatment interrogation conversational system and the intensified learning method applied to the system
CN110209791A (en) * 2019-06-12 2019-09-06 百融云创科技股份有限公司 It is a kind of to take turns dialogue intelligent speech interactive system and device more

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
QUYNH THI NGOC DO等: "Domain Adaptation in Semantic Role Labeling Using", 《IEEE》 *
卢微: "隐马尔可夫模型在自然语言理解研究中的应用", 《CNKI》 *
郑逢斌等: "特定领域中语义校对系统的开发", 《CNKI》 *

Also Published As

Publication number Publication date
CN110781685B (en) 2022-08-19

Similar Documents

Publication Publication Date Title
US11782970B2 (en) Query categorization based on image results
CN106919702B (en) Keyword pushing method and device based on document
CN108509417B (en) Title generation method and device, storage medium and server
US9898554B2 (en) Implicit question query identification
US8856124B2 (en) Co-selected image classification
US7949643B2 (en) Method and apparatus for rating user generated content in search results
CN106682147A (en) Mass data based query method and device
CN109299245B (en) Method and device for recalling knowledge points
US8996622B2 (en) Query log mining for detecting spam hosts
US8874553B2 (en) Establishing “is a” relationships for a taxonomy
CN107256267A (en) Querying method and device
US9286379B2 (en) Document quality measurement
US20100191758A1 (en) System and method for improved search relevance using proximity boosting
CN110888990A (en) Text recommendation method, device, equipment and medium
CN101155182A (en) Garbage information filtering method and apparatus based on network
CN107330079B (en) Method and device for presenting rumor splitting information based on artificial intelligence
CN111127051B (en) Multi-channel dynamic attribution method, device, server and storage medium
CN111538931A (en) Big data-based public opinion monitoring method and device, computer equipment and medium
CN103577504A (en) Method and device for putting personalized contents
US20230376531A1 (en) Media contextual information for a displayed resource
CN110889024A (en) Method and device for calculating information-related stock
CN111488453B (en) Resource grading method, device, equipment and storage medium
CN110781685B (en) Method for automatically marking correctness of semantic analysis result based on user feedback
US20100082694A1 (en) Query log mining for detecting spam-attracting queries
US20140095411A1 (en) Establishing &#34;is a&#34; relationships for a taxonomy

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant