CN117056479A - Intelligent question-answering interaction system based on semantic analysis engine - Google Patents
Intelligent question-answering interaction system based on semantic analysis engine Download PDFInfo
- Publication number
- CN117056479A CN117056479A CN202310980737.3A CN202310980737A CN117056479A CN 117056479 A CN117056479 A CN 117056479A CN 202310980737 A CN202310980737 A CN 202310980737A CN 117056479 A CN117056479 A CN 117056479A
- Authority
- CN
- China
- Prior art keywords
- question
- module
- input
- answering
- semantic
- 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.)
- Pending
Links
- 238000004458 analytical method Methods 0.000 title claims abstract description 34
- 230000003993 interaction Effects 0.000 title claims abstract description 25
- 238000012795 verification Methods 0.000 claims abstract description 18
- 238000000354 decomposition reaction Methods 0.000 claims abstract description 5
- 230000004927 fusion Effects 0.000 claims abstract description 5
- 230000006399 behavior Effects 0.000 claims description 65
- 238000000034 method Methods 0.000 claims description 57
- 230000008569 process Effects 0.000 claims description 36
- 230000006870 function Effects 0.000 claims description 19
- 238000012545 processing Methods 0.000 claims description 18
- 238000007689 inspection Methods 0.000 claims description 16
- 238000002372 labelling Methods 0.000 claims description 12
- 238000001914 filtration Methods 0.000 claims description 7
- 238000007405 data analysis Methods 0.000 claims description 6
- 238000011156 evaluation Methods 0.000 claims description 5
- 238000000605 extraction Methods 0.000 claims description 5
- 230000009471 action Effects 0.000 claims description 3
- 235000014510 cooky Nutrition 0.000 claims description 3
- 238000011049 filling Methods 0.000 claims description 3
- 230000011218 segmentation Effects 0.000 claims description 3
- 238000004891 communication Methods 0.000 abstract description 3
- 238000007726 management method Methods 0.000 description 17
- 238000003058 natural language processing Methods 0.000 description 13
- 230000006872 improvement Effects 0.000 description 6
- 238000012549 training Methods 0.000 description 6
- 238000005516 engineering process Methods 0.000 description 5
- 230000008676 import Effects 0.000 description 5
- 230000002452 interceptive effect Effects 0.000 description 4
- 238000013461 design Methods 0.000 description 3
- 238000012423 maintenance Methods 0.000 description 3
- 230000008901 benefit Effects 0.000 description 2
- 230000009466 transformation Effects 0.000 description 2
- 230000000007 visual effect Effects 0.000 description 2
- 238000013459 approach Methods 0.000 description 1
- 230000004888 barrier function Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000001419 dependent effect Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 230000014509 gene expression Effects 0.000 description 1
- 230000009191 jumping Effects 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000002035 prolonged effect Effects 0.000 description 1
- 230000004044 response Effects 0.000 description 1
- 238000004088 simulation Methods 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/33—Querying
- G06F16/332—Query formulation
- G06F16/3329—Natural language query formulation or dialogue systems
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/33—Querying
- G06F16/3331—Query processing
- G06F16/334—Query execution
- G06F16/3344—Query execution using natural language analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/33—Querying
- G06F16/3331—Query processing
- G06F16/334—Query execution
- G06F16/3346—Query execution using probabilistic model
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/35—Clustering; Classification
- G06F16/353—Clustering; Classification into predefined classes
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/36—Creation of semantic tools, e.g. ontology or thesauri
- G06F16/367—Ontology
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/30—Semantic analysis
- G06F40/35—Discourse or dialogue representation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N5/00—Computing arrangements using knowledge-based models
- G06N5/02—Knowledge representation; Symbolic representation
- G06N5/022—Knowledge engineering; Knowledge acquisition
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Databases & Information Systems (AREA)
- Artificial Intelligence (AREA)
- Mathematical Physics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Probability & Statistics with Applications (AREA)
- Human Computer Interaction (AREA)
- Animal Behavior & Ethology (AREA)
- Evolutionary Computation (AREA)
- Computing Systems (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Audiology, Speech & Language Pathology (AREA)
- General Health & Medical Sciences (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention relates to an intelligent question-answering interaction system based on a semantic analysis engine, which comprises an input module, a first text module and a second text module, wherein the input module is used for converting an external command into first text information; the pre-recognition module is used for recognizing whether the first text information is the human inquiry information or not; the semantic decomposition module is used for carrying out semantic analysis on the artificial inquiry information to obtain a semantic phrase if the first text information is the artificial inquiry information; and the query module is used for querying the knowledge graph and the question-answer engine according to the semantic phrase, fusing the query result of the knowledge graph and the query result of the question-answer engine, verifying the fusion result, and generating an answer after verification. The invention can comprehensively cover the data content of the knowledge base and help the user learn and grasp knowledge information in the related field in a question-and-answer mode. Through autonomous learning, the user intention is accurately identified, intelligent question-answering with the user is realized, the question-answering communication efficiency is high, and the user experience is good.
Description
Technical Field
The invention relates to an intelligent question-answering interaction system based on a semantic analysis engine, and belongs to the technical field of artificial intelligent question-answering of electric power professions.
Background
The method is characterized in that a knowledge management system of an integrated engineering document library, an expert library, a question library, a tag library and a standard system library is built by a unit, so that invisible knowledge in the technical knowledge field of the power profession is changed into explicit knowledge, the willingness of knowledge sharing and learning of staff of the unit is improved, and the problem that knowledge cannot be effectively deposited and inherited is solved.
1. The engineering and the power transformation profession are taken as the prior test field, personalized intelligent knowledge management application of the power professional technology is created, the service quality and efficiency of engineering accumulated knowledge experience are effectively improved, the existing intelligent knowledge base is perfected, and convenient, intelligent and personalized knowledge service capability support is provided for all project engineering and professionals.
2. And establishing knowledge point association between the professional knowledge base and the intelligent questions and answers, improving the quality of engineering knowledge retrieval service, realizing high-efficiency utilization of the prior knowledge resources and eliminating system barriers.
3. The intelligent knowledge base is promoted and implemented in the engineering and power transformation profession, so that engineering problems are better solved, engineering experience is summarized, engineering materials are integrated, engineering construction is high-quality and high-efficiency completed, a set of interactive dynamic data management center is formed, the problems of knowledge walking and experience walking are solved, knowledge and experience are solidified, and a complete knowledge system is established.
The core module for realizing the functions is an intelligent question-answering robot, but the current intelligent question-answering robot is not special for a professional knowledge base (knowledge graph) in the electric power field, and related robots are mostly concentrated on customer service robots or automatic answer robots similar to trigger keywords, and are relatively mechanical, inaccurate and cannot be used for semantic analysis.
For this purpose, we propose an intelligent question-answer interaction system based on a semantic parsing engine.
Disclosure of Invention
Aiming at the defects existing in the prior art, the invention provides an intelligent question-answer interaction system based on a semantic analysis engine, which comprises the following specific technical scheme:
an intelligent question-answering interaction system based on a semantic analysis engine comprises:
the input module is used for converting an external command into first text information;
the pre-recognition module is used for recognizing whether the first text information is the human inquiry information or not;
the semantic decomposition module is used for carrying out semantic analysis on the artificial inquiry information to obtain a semantic phrase if the first text information is the artificial inquiry information;
the query module queries the knowledge graph and the question-answer engine according to the semantic phrase, fuses the query result of the knowledge graph and the query result of the question-answer engine, verifies the fusion result, and generates an answer if the verification is passed;
the knowledge graph is called according to the semantic phrase, and knowledge graph inquiry and logic reasoning are carried out in the knowledge graph, so that an inquiry result of the knowledge graph is obtained;
the semantic phrase is transferred to a question-answering engine through an access layer, and the question-answering engine generates a query result of the question-answering engine through NLP processing, intention recognition, text retrieval and model prediction processing.
Further improvements, the data layer of the question-answering engine comprises question-answering data and knowledge base data, and an intention recognition model and a question-answering model are generated according to the data layer data; the operation management of the data layer is realized through a labeling module, and the labeling module is oriented to editors and supports evaluation labeling and knowledge base expansion labeling of on-line problems.
In a further improvement, the logic layer of the question-answering engine is an NLU module for carrying out NLP processing on the semantic phrases, the function of the search module of the question-answering engine is to realize the semantic matching function based on the Bi-LSTM algorithm and the DSSM model by the search matching function based on the BM25 model, and the rule matching module of the question-answering engine is combined with the offline experimental rule to realize the rule matching on the semantic phrases.
Further improved, the NLU module in the logic layer of the question-answering engine performs NLP processing on the semantic phrases, including one or more of word segmentation, part-of-speech analysis, synonym and keyword extraction and part-of-speech filtering.
And in a further improvement, the rule matching module, the searching module and the predicting module are respectively processed in parallel to obtain processing sub-results, and the sub-results of the rule matching module, the searching module and the predicting module are fused and sequenced according to a preset strategy and priority to obtain a final recognition result.
Further improved, the data layer of the question-answering engine also comprises an NLU database and a multi-round question-answering database, the logic layer of the question-answering engine also comprises a dialogue management module, the dialogue management module comprises a dialogue state tracking model, a dialogue strategy learning module and a natural language generating module, and after the semantic phrase reaches the logic layer, the semantic phrase and the answer are subjected to intention matching through a rule matching module; and for the matched and hit semantic phrases, performing Bi-LSTM algorithm, CRF model and word slot joint recognition through an NLU module, after the recognition result is disambiguated through a knowledge graph and acquired by related entities, generating a state automatic file by a dialogue management module, matching a group of reply strategies by a dialogue state tracking model based on the state automatic file and updating a user state, selecting an answer template by a dialogue strategy learning module based on the user state, and finally filling information into the answer template through a natural language generation module to generate a query result of a question-answer engine.
The system is further improved, and further comprises an intelligent recommendation module, wherein the question and answer engine recommends the user of possible questions according to the user identity, the user history behavior and the user preference information; when a user inputs a problem, the intelligent recommendation module guides and corrects the input of the user according to the input content, so that the user is helped to quickly locate the problem.
In a further improvement, the intelligent recommendation process of the intelligent recommendation module is divided into a recall process and a reordering process, wherein the recall process obtains content possibly interested by a user, and the ordering process performs scoring, ordering and filtering on the recalled content so as to obtain a recommendation list which accords with the idea of the user to input.
In a further improvement, human-computer verification is performed on the input behaviors of the user through the pre-recognition module, and a multidimensional data analysis model is constructed through behavior data, equipment characteristics and network data of the user, wherein the multidimensional data analysis model comprises one or more of browser characteristic inspection, mouse behavior inspection, page window inspection, cookie inspection and input method behavior inspection.
Further improvement, in the input method behavior checking process, comprises matching input habit with input habit in sample data, misinput behavior verification, pre-input behavior verification,
the pre-input behavior verification refers to:
before outputting the pre-input content as the first text information through the input module, checking whether the pre-input content has the recorded behavior but is not finally outputted as the first text information;
checking whether a deleting or correcting action exists in the recording process;
performing similarity check on the text entity in the input process and the finally output first text information, judging correct behavior input if the check result is greater than or equal to a first threshold value, counting the input probability of the correct behavior input, and judging artificial input if the input probability is greater than or equal to a second threshold value;
and performing similarity check on the deleted content in the input process and the finally output first text information, classifying suspected artificial input behaviors if the check result is smaller than or equal to a third threshold value, counting the suspected probability of the suspected artificial input behaviors, and judging that the input is artificial when the suspected probability is larger than or equal to a fourth threshold value.
The invention has the beneficial effects that:
1. the intelligent question-answering interaction system based on the semantic analysis engine is an intelligent interactive question-answering system with natural language understanding capability, can comprehensively cover the data content of a knowledge base, and helps a user learn knowledge information in related fields through question-answering modes.
2. The intelligent question-answering interaction system based on the semantic analysis engine provided by the invention accurately identifies the user intention through autonomous learning, realizes intelligent question-answering with the user, and has the advantages of high question-answering communication efficiency and good user experience.
Detailed Description
The present invention will be described in further detail with reference to the following examples in order to make the objects, technical solutions and advantages of the present invention more apparent. It should be understood that the specific embodiments described herein are for purposes of illustration only and are not intended to limit the scope of the invention.
Example 1
The intelligent knowledge management application of the power professional technology is built in the unit, the module functions of knowing, knowledge searching and inquiring, and realizing personalized knowledge management and display aiming at engineering are developed in the unit, wherein the application module comprises the engineering personalized intelligent dialogue management, engineering problem tracking and experience sharing, visual knowledge graph, staff learning training, data statistics analysis, personal center and other module functions, and the application module comprises the design, development and implementation work of the engineering personalized intelligent dialogue management, engineering problem tracking and experience sharing. The specific functions are as follows:
1. intelligent dialog management
The system supports the robot to support webpage and API interface channel access.
2. Knowledge base management
The knowledge base dialog system is capable of entering knowledge data related to business question-answer knowledge into the system. Support new, modified, deleted, one-key training, release. And supporting knowledge dialogue batch import and one-key knowledge export.
3. Knowledge dialogue batch import
The knowledge dialogue system supports the user to import question pair data from local batch, import question and answer to know standard questions and corresponding similar question and answer, answer corresponding to questions and content classification of knowledge question and answer.
4. Dialogue association question
The user may set one or more sets of associated question-answer knowledge for each question of the knowledge dialog system. The function can help the user group question-answer pairs in the knowledge base for business processes or question intentions, and the intelligent dialogue system can show the question-answer knowledge in the same scope of the agreement graph to the user through question-answer association questions.
5. Dialogue knowledge assessment
Each knowledge session supports a session knowledge result evaluation function that an end user can evaluate based on the performance of the session system. After the knowledge question-answering system returns answers in matching according to questions presented by the terminal user, the terminal user can manually mark the contents of the returned answers, and the system collects feedback and use conditions of the terminal user.
6. Professional word stock
The professional word stock allows a user to customize a professional word stock for the knowledge base intelligent dialogue system according to the power industry, and the content in the professional word stock is used for analyzing questions and answers. The user can override the user's more possibilities for problem description through the setting of the professional word. The user can add, import and delete the transformer trade technical nouns in batches to set the dictionary, and the robot can answer the professional questions more accurately.
7. Semantic question-answer similarity analysis
And carrying out similarity analysis on the knowledge base and the user questions and completing answers, and supporting variable semantic generalization understanding capability. The similarity analysis model is compatible with various corpuses and has the characteristics of high precision and less time consumption. Supporting dialogue association questions and knowledge evaluation.
8. Multiple rounds of dialogue
The multi-round dialogue function is based on deep semantic analysis service, combines a context-free grammar writing and analysis system, carries out deep understanding on dialogue intention and topic content of a user, and finds dialogue context logic association and content association. Multiple rounds of conversations may be configured and automatically handled to complete the correct reply in conjunction with multiple rounds of content.
9. User intent analysis and key information extraction
Based on the semantic parsing structure and the user configuration template, user intention is accurately analyzed, key information in user questions is identified, and different query answers are completed based on the information.
10. Dialog flow design
The dialogue flow is based on deep semantic understanding and dialogue central control, services such as knowledge base questions and answers, intention dialogue templates and the like are integrated, and a user draws a dialogue flow chart according to business scene requirements through visual dialogue flow node setting. The design flow supports operations such as dialog graph nesting, dialog path jumping, backtracking and circulation, and the like, and simultaneously supports the calling of a third party service, codes are not needed, only one-key training is needed to be effective in real time, and a high-quality intelligent service virtual assistant is provided for clients.
11. Session flow diagnostics
The background of the conversation flow system can count conversation paths which a user walks in the process of interacting with the intelligent service assistant, and judge the accessed times of different conversation nodes in the conversation flow chart, and the execution times of each conversation flow line.
12. Dialogue welcome language
The user can guide corpus for the user-defined service of the dialogue system, each time when the user interacts with the dialogue system, the system can actively call the user, introduce the capability range of the dialogue system to the user, guide the user to start dialogue interaction, and actively push the problem that the client may consult.
13. Dialogue assistance guidance
Scenarios for triggering help functions include active departure of the user, inability of the dialog system to understand the problems of the end user, and prolonged non-response of the end user.
14. Handling page turns
In the process of the user talking with the robot, the user can support the document, the link or the button to send, and the user can click the document to preview or skip the interface.
15. Sensitive word stock
Setting a sensitive word library for defining business or general sensitive words, which are similar to political sensitive words, complaint sensitive words, non-civilized expressions and the like;
when the user question method triggers the sensitive word, special processing or replying can be carried out according to the configured strategy.
16. Robot trainer
And the method supports the convergence of high-frequency difficult problems according to historical dialogue data, supports iteration and retraining, enhances the capability of an intelligent customer service model and creates a full closed loop. The robot can utilize the historical information and the man-machine interaction historical record to carry out intelligent knowledge summarization analysis, comb out corresponding knowledge points, similar question methods and new knowledge points, and recommend the knowledge points to maintenance personnel to carry out knowledge base maintenance. Meanwhile, correct reply to the knowledge base content can be completed through manual participation analysis.
17. Robot reading and understanding
And the documents, specifications and the like of the power industry uploaded by the machine reading system user are pre-trained on the massive marked texts or marked data in the early stage to provide intelligent answers and reading of the problems of specific document contents for the user.
18. Intelligent question and answer
The intelligent interactive knowledge question and answer assistant is used for reading, understanding and learning aiming at a background question and answer library, and an intelligent interactive question and answer system with natural language understanding capability is formed by combining background question and answer library data. The knowledge base data content can be covered in an omnibearing way, and a user is helped to learn knowledge information in the related field through a question-answer mode.
19. Robot assistant
The method can support the creation of a plurality of robot virtual assistants, similar equipment maintenance assistants, different robot assistants, different knowledge bases and services oriented to different scenes.
The intelligent question-answering interaction system based on the semantic analysis engine is called an intelligent question-answering robot for short, and is one of core functions in intelligent knowledge management application of power professional technology.
Example 2
An intelligent question-answering interaction system based on a semantic analysis engine comprises,
the input module is used for converting an external command into first text information; such as various input method software, voice input text conversion, etc. The external command may be text or voice, etc.
The pre-recognition module is used for recognizing whether the first text information is the human inquiry information or not; mainly filtering meaningless characters to a greater extent, such as misoperation; or to identify human-machine behavior.
The semantic decomposition module is used for carrying out semantic analysis on the artificial inquiry information to obtain a semantic phrase if the first text information is the artificial inquiry information;
the query module queries the knowledge graph and the question-answer engine according to the semantic phrase, fuses the query result of the knowledge graph and the query result of the question-answer engine, verifies the fusion result, and generates an answer if the verification is passed;
the knowledge graph is called according to the semantic phrase, and knowledge graph inquiry and logic reasoning are carried out in the knowledge graph, so that an inquiry result of the knowledge graph is obtained;
the semantic phrase is transferred to a question-answering engine through an access layer, and the question-answering engine generates a query result of the question-answering engine through NLP processing, intention recognition, text retrieval and model prediction processing.
Example 3
In the embodiment 2, the data layer of the question and answer engine in 2 comprises question and answer data and knowledge base data, and an intention recognition model and a question and answer model are generated according to the data layer data; the operation management of the data layer is realized through a labeling module, and the labeling module is oriented to editors and supports evaluation labeling and knowledge base expansion labeling of on-line problems.
Example 4
In embodiment 2, the logic layer of the question-answering engine is an NLU module for performing NLP processing on the semantic phrase, the function of the search module of the question-answering engine is to realize the semantic matching function based on the Bi-LSTM algorithm and the DSSM model (deep semantic matching model) through the search matching function based on the BM25 model, and the rule matching module of the question-answering engine combines with the offline experimental rule to realize rule matching on the semantic phrase.
The NLU module in the logic layer of the question-answering engine carries out NLP processing on the semantic phrases, wherein the NLP processing comprises one or more of word segmentation, part-of-speech analysis, synonym and keyword extraction and part-of-speech filtering.
Natural language processing (Natural Language Processing, NLP for short), natural language understanding (Natural Language Understanding, NLU for short) and natural language generation (Natural Language Generation, NLG for short), which are another core task of natural speech processing (NLP) for the main purpose of reducing the gap between human and machine communication, and converting data in non-language format into language format that can be understood by human.
The intelligent question-answering interaction system based on the semantic analysis engine is called an intelligent question-answering robot for short, and the question-answering flow is as follows:
the user inputs a question on a webpage, a PC client or a mobile client, and an external command (question) is converted into first text information through an input module; the pre-recognition module is used for recognizing whether the first text information is human query information or not, so that misoperation (including but not limited to whether input is interrupted, nonsensical characters, character strings and the like), machine operation, correct human operation and the like are distinguished, and the main function is to distinguish normal human operation from machine operation and intercept malicious behaviors.
If the first text information is the human query information, carrying out semantic analysis on the human query information through a semantic decomposition module to obtain a semantic phrase;
the query module queries the knowledge graph and the question-answering engine according to the semantic phrase, invokes the knowledge graph according to the semantic phrase, and performs knowledge graph query and logic reasoning in the knowledge graph to obtain a query result of the knowledge graph; and fusing the query result of the knowledge graph and the query result of the question-answer engine, verifying the fusion result, and generating an answer after verification.
The semantic phrase is transferred to a question and answer engine through an access layer, and the question and answer engine generates a query result of the question and answer engine through NLP (natural language processing technology), intention recognition, text retrieval, model prediction and other processes.
Example 5
In embodiment 4, the rule matching module, the searching module and the predicting module respectively process in parallel to obtain processing sub-results, and the sub-results of the rule matching module, the searching module and the predicting module are fused and sequenced according to a preset strategy and priority to obtain a final recognition result. So designed, the time consumption of the question-answer process is only dependent on the time consumption corresponding to the maximum time consumption module.
Example 6
In embodiment 2, the data layer of the question-answering engine further includes an NLU database and a multi-round question-answering database, the logic layer of the question-answering engine further includes a dialogue management module, the dialogue management module includes a dialogue state tracking model (DST), a dialogue strategy learning module (DPL), and a natural language generation module (NLG), and after the semantic phrase arrives at the logic layer, the semantic phrase (question) and the answer are subjected to intention matching through a rule matching module; and for the matched and hit semantic phrases, performing Bi-LSTM algorithm, CRF model and word slot joint recognition through an NLU module, after the recognition result is disambiguated through a knowledge graph and acquired by related entities, generating a state automatic file by a dialogue management module, matching a group of reply strategies by a dialogue state tracking model based on the state automatic file and updating a user state, selecting an answer template by a dialogue strategy learning module based on the user state, and finally filling information into the answer template through a natural language generation module to generate a query result of a question-answer engine.
Example 7
The intelligent question-answering interaction system based on the semantic analysis engine further comprises an intelligent recommendation module, wherein when a user enters an intelligent question-answering page, the question-answering engine recommends a question which can be asked for the user according to the user identity, the user history behavior and the user preference information collected by a third party; when a user inputs a problem, the intelligent recommendation module guides and corrects the input of the user according to the input content, so that the user is helped to quickly locate the problem.
The intelligent recommendation module realizes intelligent operations such as inputting association, guessing you want to ask, and the like, quickly corrects and guides users to ask questions, and achieves the effects of reducing input errors and realizing problem prejudgement.
Example 8
The intelligent recommendation process of the intelligent recommendation module is divided into a recall process and a reordering process, wherein the recall process obtains content possibly interested by a user, and the ordering process performs scoring ordering filtering on the recalled content so as to obtain a recommendation list which accords with the idea of the user to input.
In the intelligent recommendation module, the answer accuracy and the question resolution of the intelligent question answering robot are kept at a high level, and the practicability is high.
Example 9
In embodiment 2, human-computer verification is performed on the input behavior of the user through the pre-recognition module, and a multidimensional data analysis model is constructed through behavior data, device characteristics and network data of the user, wherein the multidimensional data analysis model comprises one or more of browser characteristic inspection, mouse behavior inspection, page window inspection, cookie inspection and input method behavior inspection.
All browsers have differences, and the authenticity of the browser environment can be checked by various front-end related means. Mouse behaviors such as clicking, double clicking, moving, etc., are obviously different from the habit and track of a human.
And (3) establishing a human behavior model through analysis of the collected multidimensional data so as to judge whether the user is a robot.
In addition, in the input method behavior checking process, the input method behavior checking method comprises matching whether the input habit is matched with the input habit in the sample data (collected by a third party), misinput behavior verification and pre-input behavior verification,
the pre-input behavior verification refers to:
before outputting the pre-input content as the first text information through the input module, checking whether the pre-input content has the recorded behavior but is not finally outputted as the first text information; that is, if the user inputs by using the input method, since the user needs to think in the retrieval process, the behavior which is input but is not output as the first text information means that the user has input a certain period of information by using the input method, but does not input the information into the input module, and if a large number of such behaviors exist, the behavior is not made by the robot.
Checking whether a deleting or correcting action exists in the recording process; similarly, if there are a large number of such behaviors, it is stated that the behavior is not that of the robot.
Performing similarity check on the text entity in the input process and the finally output first text information, judging correct behavior input if the check result is greater than or equal to a first threshold value, counting the input probability of the correct behavior input, and judging artificial input if the input probability is greater than or equal to a second threshold value; the behavior is determined not to be the robot input by only judging whether the behavior is the correct behavior input or not by calling the database permission related to the input method, and the AI robot is simulated to break through the behavior; however, if the second threshold is optimized by analysis training by gathering data, only less than 26% of the probability is simulated and breached by the AI robot.
Carrying out randomness check on the frequency in the input process; robots are often subject to a large number of trial and error in a short time, exhibit a very pronounced regularity, through which inspection some of the apparent regular robot behaviour can be masked.
And performing similarity check on the deleted content in the input process and the finally output first text information, classifying suspected artificial input behaviors if the check result is smaller than or equal to a third threshold value, counting the suspected probability of the suspected artificial input behaviors, and judging that the input is artificial when the suspected probability is larger than or equal to a fourth threshold value. That is, taking the input method as an example, in the input process, the input method continuously deletes and organizes sentences so as to continuously approach the final first text information, and by checking the behaviors, a large number of simulation behaviors of the AI robot can be greatly shielded, and the AI robot can be continuously approached only after a large number of training; similarly, the behavior needs to depend on invoking the database permission related to the input method, if the behavior is judged to be not the robot input only by whether the behavior is the suspected human input behavior, the behavior is simulated by the AI robot with the probability of exceeding 41 percent; however, if analysis training is performed by collecting data, a fourth threshold is optimized, and the fourth threshold is an interval value, and the accuracy is affected by too large or too small; therefore, if the probability of the suspected probability finally falls within the interval of the fourth threshold, only the probability of less than 9% is simulated by the AI robot.
The AI robot in this embodiment refers to a verification code recognition technology based on machine learning.
The foregoing description of the preferred embodiments of the invention is not intended to be limiting, but rather is intended to cover all modifications, equivalents, and alternatives falling within the spirit and principles of the invention.
Claims (10)
1. The intelligent question-answering interaction system based on the semantic analysis engine is characterized by comprising the following steps:
the input module is used for converting an external command into first text information;
the pre-recognition module is used for recognizing whether the first text information is the human inquiry information or not;
the semantic decomposition module is used for carrying out semantic analysis on the artificial inquiry information to obtain a semantic phrase if the first text information is the artificial inquiry information;
the query module queries the knowledge graph and the question-answer engine according to the semantic phrase, fuses the query result of the knowledge graph and the query result of the question-answer engine, verifies the fusion result, and generates an answer if the verification is passed;
the knowledge graph is called according to the semantic phrase, and knowledge graph inquiry and logic reasoning are carried out in the knowledge graph, so that an inquiry result of the knowledge graph is obtained;
the semantic phrase is transferred to a question-answering engine through an access layer, and the question-answering engine generates a query result of the question-answering engine through NLP processing, intention recognition, text retrieval and model prediction processing.
2. The intelligent question-answering interaction system based on a semantic parsing engine according to claim 1, wherein: the data layer of the question-answering engine comprises question-answering data and knowledge base data, and an intention recognition model and a question-answering model are generated depending on the data layer data; the operation management of the data layer is realized through a labeling module, and the labeling module is oriented to editors and supports evaluation labeling and knowledge base expansion labeling of on-line problems.
3. The intelligent question-answering interaction system based on a semantic parsing engine according to claim 1, wherein: the logic layer of the question-answering engine is that an NLU module carries out NLP processing on semantic phrases, and the function of a search module of the question-answering engine aims at the semantic phrases, and the prediction module of the question-answering engine realizes the semantic matching function based on a Bi-LSTM algorithm and a DSSM model by the search matching function based on a BM25 model, and the rule matching module of the question-answering engine combines with offline experimental rules to realize rule matching on the semantic phrases.
4. The intelligent question-answering interaction system based on a semantic parsing engine according to claim 3, wherein: NLU module in the logic layer of question-answering engine carries out NLP processing on semantic phrase including one or several of word segmentation, part-of-speech analysis, synonym extraction and keyword extraction and part-of-speech filtering.
5. The intelligent question-answering interaction system based on a semantic parsing engine according to claim 3, wherein: and the rule matching module, the retrieval module and the prediction module are respectively processed in parallel to obtain processing sub-results, and the sub-results of the rule matching module, the retrieval module and the prediction module are fused and sequenced according to a preset strategy and priority to obtain a final recognition result.
6. The intelligent question-answering interaction system based on a semantic parsing engine according to claim 3, wherein: the data layer of the question-answering engine also comprises an NLU database and a multi-round question-answering database, the logic layer of the question-answering engine also comprises a dialogue management module, the dialogue management module comprises a dialogue state tracking model, a dialogue strategy learning module and a natural language generating module, and after the semantic phrase reaches the logic layer, the semantic phrase and the answer are subjected to intention matching through a rule matching module; and for the matched and hit semantic phrases, performing Bi-LSTM algorithm, CRF model and word slot joint recognition through an NLU module, after the recognition result is disambiguated through a knowledge graph and acquired by related entities, generating a state automatic file by a dialogue management module, matching a group of reply strategies by a dialogue state tracking model based on the state automatic file and updating a user state, selecting an answer template by a dialogue strategy learning module based on the user state, and finally filling information into the answer template through a natural language generation module to generate a query result of a question-answer engine.
7. The intelligent question-answering interaction system based on a semantic parsing engine according to claim 1, wherein: the intelligent recommendation module is used for recommending possible questioning questions to the user according to the user identity, the user history behavior and the user preference information; when a user inputs a problem, the intelligent recommendation module guides and corrects the input of the user according to the input content, so that the user is helped to quickly locate the problem.
8. The intelligent question-answering interaction system based on a semantic parsing engine according to claim 7, wherein: the intelligent recommendation process of the intelligent recommendation module is divided into a recall process and a reordering process, wherein the recall process obtains content possibly interested by a user, and the ordering process performs scoring ordering filtering on the recalled content so as to obtain a recommendation list which accords with the idea of the user to input.
9. The intelligent question-answering interaction system based on a semantic parsing engine according to claim 1, wherein: and carrying out man-machine verification on the input behaviors of the user through the pre-recognition module, and constructing a multidimensional data analysis model through behavior data, equipment characteristics and network data of the user, wherein the multidimensional data analysis model comprises one or more of browser characteristic inspection, mouse behavior inspection, page window inspection, cookie inspection and input method behavior inspection.
10. The intelligent question-answering interaction system based on a semantic parsing engine according to claim 9, wherein: in the input method behavior checking process, the input method behavior checking method comprises matching whether the input habit is matched with the input habit in the sample data, misinput behavior verification and pre-input behavior verification,
the pre-input behavior verification refers to:
before outputting the pre-input content as the first text information through the input module, checking whether the pre-input content has the recorded behavior but is not finally outputted as the first text information;
checking whether a deleting or correcting action exists in the recording process;
performing similarity check on the text entity in the input process and the finally output first text information, judging correct behavior input if the check result is greater than or equal to a first threshold value, counting the input probability of the correct behavior input, and judging artificial input if the input probability is greater than or equal to a second threshold value;
and performing similarity check on the deleted content in the input process and the finally output first text information, classifying suspected artificial input behaviors if the check result is smaller than or equal to a third threshold value, counting the suspected probability of the suspected artificial input behaviors, and judging that the input is artificial when the suspected probability is larger than or equal to a fourth threshold value.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310980737.3A CN117056479A (en) | 2023-08-07 | 2023-08-07 | Intelligent question-answering interaction system based on semantic analysis engine |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310980737.3A CN117056479A (en) | 2023-08-07 | 2023-08-07 | Intelligent question-answering interaction system based on semantic analysis engine |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN117056479A true CN117056479A (en) | 2023-11-14 |
Family
ID=88668507
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310980737.3A Pending CN117056479A (en) | 2023-08-07 | 2023-08-07 | Intelligent question-answering interaction system based on semantic analysis engine |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117056479A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117407514A (en) * | 2023-11-28 | 2024-01-16 | 星环信息科技(上海)股份有限公司 | Solution plan generation method, device, equipment and storage medium |
| CN117592489A (en) * | 2023-11-30 | 2024-02-23 | 北京快牛智营科技有限公司 | Method and system for realizing electronic commerce commodity information interaction by using large language model |
| CN117952022A (en) * | 2024-03-26 | 2024-04-30 | 杭州广立微电子股份有限公司 | Yield multi-dimensional interactive system, method, computer equipment and storage medium |
-
2023
- 2023-08-07 CN CN202310980737.3A patent/CN117056479A/en active Pending
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117407514A (en) * | 2023-11-28 | 2024-01-16 | 星环信息科技(上海)股份有限公司 | Solution plan generation method, device, equipment and storage medium |
| CN117592489A (en) * | 2023-11-30 | 2024-02-23 | 北京快牛智营科技有限公司 | Method and system for realizing electronic commerce commodity information interaction by using large language model |
| CN117592489B (en) * | 2023-11-30 | 2024-05-17 | 北京快牛智营科技有限公司 | Method and system for realizing electronic commerce commodity information interaction by using large language model |
| CN117952022A (en) * | 2024-03-26 | 2024-04-30 | 杭州广立微电子股份有限公司 | Yield multi-dimensional interactive system, method, computer equipment and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111046132B (en) | Customer service question-answering processing method and system for searching multiple rounds of conversations | |
| CN114547329A (en) | Method for establishing pre-training language model, semantic analysis method and device | |
| CN117056479A (en) | Intelligent question-answering interaction system based on semantic analysis engine | |
| KR100818979B1 (en) | Dialog management apparatus and method for chatting agent | |
| KR102041621B1 (en) | System for providing artificial intelligence based dialogue type corpus analyze service, and building method therefor | |
| CN110765277B (en) | Knowledge-graph-based mobile terminal online equipment fault diagnosis method | |
| CN112527998A (en) | Reply recommendation method, reply recommendation device and intelligent device | |
| CN111026840B (en) | Text processing method, device, server and storage medium | |
| CN113672708A (en) | Language model training method, question and answer pair generation method, device and equipment | |
| Lommatzsch et al. | An Information Retrieval-based Approach for Building Intuitive Chatbots for Large Knowledge Bases. | |
| CN112417127A (en) | Method, device, equipment and medium for training conversation model and generating conversation | |
| CN110597968A (en) | Reply selection method and device | |
| CN117520522B (en) | Intelligent dialogue method and device based on combination of RPA and AI and electronic equipment | |
| CN114186040A (en) | Operation method of intelligent robot customer service | |
| CN116450799B (en) | Intelligent dialogue method and equipment applied to traffic management service | |
| CN117828024A (en) | Plug-in retrieval method, device, storage medium and equipment | |
| Hong et al. | Comprehensive technology function product matrix for intelligent chatbot patent mining | |
| CN116186220A (en) | Information retrieval method, question and answer processing method, information retrieval device and system | |
| CN114661864A (en) | Psychological consultation method and device based on controlled text generation and terminal equipment | |
| CN114969347A (en) | Defect duplication checking implementation method and device, terminal equipment and storage medium | |
| CN111753540B (en) | Method and system for collecting text data to perform Natural Language Processing (NLP) | |
| CN114942981A (en) | Question-answer query method and device, electronic equipment and computer readable storage medium | |
| WO2019103569A1 (en) | Method for improving performance of voice recognition on basis of context, computer apparatus, and computer-readable recording medium | |
| CN111460106A (en) | Information interaction method, device and equipment | |
| CN114154513B (en) | Automatic domain semantic web construction method and system |
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 |