CN116701579A - Information reply system, method and computer readable storage medium - Google Patents
Information reply system, method and computer readable storage medium Download PDFInfo
- Publication number
- CN116701579A CN116701579A CN202310147714.4A CN202310147714A CN116701579A CN 116701579 A CN116701579 A CN 116701579A CN 202310147714 A CN202310147714 A CN 202310147714A CN 116701579 A CN116701579 A CN 116701579A
- Authority
- CN
- China
- Prior art keywords
- question
- answer
- module
- answering module
- target
- 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
- 238000000034 method Methods 0.000 title claims abstract description 88
- 238000012545 processing Methods 0.000 claims abstract description 77
- 230000008569 process Effects 0.000 claims abstract description 52
- 230000011218 segmentation Effects 0.000 claims description 31
- 238000013145 classification model Methods 0.000 claims description 10
- 230000002159 abnormal effect Effects 0.000 claims description 6
- 238000004590 computer program Methods 0.000 claims description 5
- 230000000875 corresponding effect Effects 0.000 description 38
- 230000006870 function Effects 0.000 description 15
- 238000012549 training Methods 0.000 description 12
- 238000010586 diagram Methods 0.000 description 9
- 239000013598 vector Substances 0.000 description 7
- 238000004364 calculation method Methods 0.000 description 6
- 230000004044 response Effects 0.000 description 6
- 238000005516 engineering process Methods 0.000 description 5
- 230000001105 regulatory effect Effects 0.000 description 4
- 230000009471 action Effects 0.000 description 3
- 238000010276 construction Methods 0.000 description 3
- 238000000605 extraction Methods 0.000 description 3
- 238000012360 testing method Methods 0.000 description 3
- 238000012795 verification Methods 0.000 description 3
- 238000004458 analytical method Methods 0.000 description 2
- 238000004422 calculation algorithm Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000002829 reductive effect Effects 0.000 description 2
- 230000002411 adverse Effects 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 239000012634 fragment Substances 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 230000002452 interceptive effect Effects 0.000 description 1
- 238000002372 labelling Methods 0.000 description 1
- 230000000670 limiting effect Effects 0.000 description 1
- 239000004973 liquid crystal related substance Substances 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 230000002441 reversible effect Effects 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 238000010200 validation analysis 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
-
- 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
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D10/00—Energy efficient computing, e.g. low power processors, power management or thermal management
Abstract
The application provides an information reply system, an information reply method and a computer readable storage medium, and relates to the technical field of information question and answer. The system comprises: the system comprises an input interface, a serial question-answering module, a retrieval module and an output interface; the input interface is connected with the serial question-answering module, the output interface is connected with the serial question-answering module and the retrieval module, and the serial question-answering module is connected with the retrieval module; the input interface is used for receiving target information needing to be replied; the serial question-answering module comprises a first question-answering module, a second question-answering module and a third question-answering module which are connected in series according to a preset sequence and are used for processing target information to obtain target answers; the retrieval module is used for processing the target information to obtain a retrieval answer when the target information is not obtained after the serial question-answering module processes the target information; the output interface is used for outputting target answers or search answers, and can effectively improve the answer rate of the information questions and the accuracy in answer.
Description
Technical Field
The present application relates to the field of information question answering technology, and in particular, to an information answer system, method, and computer readable storage medium.
Background
Many industries and fields require learning and mastering of relevant legal and regulatory knowledge, and continuously raise their legal awareness.
In the prior art, when learning and mastering relevant legal and legal knowledge, a conventional search engine is generally used to input relevant keywords for searching so as to obtain corresponding search results for learning. However, because of the numerous laws and regulations in existence and the complexity of various laws and regulations, it is difficult for users to accurately and rapidly obtain accurate contents of related laws and regulations when searching by using a conventional search engine. The search capability of the current search engine for related problems of legal and legal information is poor, and the learning process of the legal and legal information can be adversely affected.
Disclosure of Invention
In view of the foregoing, an objective of the embodiments of the present application is to provide an information reply system, method and computer readable storage medium for improving the problem of poor searching ability of related problems of legal and regulatory information in the prior art.
To solve the above problems, in a first aspect, an embodiment of the present application provides an information reply system, including: the system comprises an input interface, a serial question-answering module, a retrieval module and an output interface;
The input interface is connected with the serial question-answering module, the output interface is connected with the serial question-answering module and the retrieval module, and the serial question-answering module is connected with the retrieval module;
the input interface is used for receiving target information to be replied;
the serial question-answering module comprises a first question-answering module, a second question-answering module and a third question-answering module which are connected in series according to a preset sequence, wherein the serial question-answering module is used for processing the target information to obtain a target answer;
the retrieval module is used for processing the target information to obtain a retrieval answer when the target information is not obtained after the serial question-answering module processes the target information;
the output interface is used for outputting the target answer or the search answer.
In the implementation process, the application applies the question-answering modules with various functions of intelligent question-answering to the information answer, and connects the question-answering modules with various functions in series so as to process the target information which is proposed by the user and needs to be answered, and obtain the corresponding target answer for replying. And when the serial question-answering module cannot answer the target information, the connected search module is used for processing the target information continuously to obtain a corresponding search answer for replying. Through the serial connection and the connection of the question-answering modules with different functions and the retrieval module, the target information can be correspondingly processed based on a plurality of different question-answering and retrieval technologies, the answer rate of the target information and the accuracy of the output answer are improved, so that a user can quickly and accurately acquire relevant legal and legal knowledge during learning, the use experience of the user is optimized, and the learning efficiency of the user is improved.
Optionally, the first question answering module includes a common question set question answering module, the second question answering module includes a knowledge graph question answering module, and the third question answering module includes a machine reading understanding question answering module.
In the implementation process, in order to further improve the efficiency and the answer rate when the serial question-answering modules generate the target answers, the serial sequence of the question-answering modules with three different functions can be limited, the common question-set question-answering module with the highest processing speed is used as the first question-answering module for processing the target information, a knowledge graph is constructed, the knowledge graph question-answering module for retrieving answers from the knowledge base of the graph is used as the second question-answering module for processing the target information, and the machine-readable understanding question-answering module capable of retrieving and extracting a plurality of unstructured text data to obtain the answers is used as the last question-answering module for processing the target information, so that the target information can be sequentially processed from multiple aspects such as the processing speed, the range of the knowledge base during retrieval, the type of the processed target information and the like, and the efficiency and the answer rate when the target answers are generated are effectively improved.
Optionally, the first question-answering module is used for judging whether the target information is normal or not;
if the target information is abnormal, the first question-answering module is further used for feeding back a re-input request;
if the target information is normal, the first question-answer module is further used for determining whether standard information with similarity higher than a similarity threshold value with the target information exists in a preset question-answer database or not based on a classification model; wherein, the question-answer database comprises a plurality of information templates;
and if the standard information is in the question-answer database, the first question-answer module is further used for taking the standard answer of the standard information as the target answer.
In the implementation process, when the first question-answer module processes the target information, whether the target information is normal or not can be judged, when the first question-answer module processes the target information according to the classification model continuously, templates related to the target information are matched in a plurality of information templates in a preset question-answer database, and when standard information with similarity higher than a similarity threshold value with the target information is provided in the question-answer database, standard answers corresponding to the standard information are used as target answers to be output. The target information can be quickly and accurately matched with whether the corresponding target answer exists or not through accurate processing of the classification model, and efficiency and accuracy of the first question-answering module in answering are effectively improved.
Optionally, the system further comprises: the updating module is used for detecting the target answer or the search answer corresponding to the target information;
if the target answer or the search answer corresponding to the target information is wrong, the updating module is specifically configured to obtain a correct answer, and add the target information and the correct answer to the question-answer database of the first question-answer module.
In the implementation process, in order to improve accuracy of the question reply, the update module may be configured to detect accuracy of the target answer or the search answer, and when detecting that the target answer or the search answer is wrong, the update module may be configured to obtain a corresponding correct answer, and add the target information and the associated correct answer to the question-reply database of the first question-reply module. The method and the device can collect and store the information and the answers of the incorrect answers, so that the first question-answer module is repeatedly updated continuously, the question-answer database of the first question-answer module is perfected continuously, and the answer rate and the accuracy of the target information when being processed in the first question-answer module are further improved.
Optionally, the second question-answering module is used for performing word segmentation on the target information to obtain first word segmentation data; traversing a plurality of nodes of a preset knowledge graph based on the first word segmentation data, and determining whether target nodes with correlation with the first word segmentation data higher than a correlation threshold value exist in the plurality of nodes;
And if the target node exists in the plurality of nodes, the second question-answering module is further used for taking the node answer corresponding to the target node as the target answer.
In the implementation process, when the second question-answering module processes the target information, in order to improve accuracy in processing, word segmentation processing can be performed on the target information first, a corresponding knowledge graph is created in advance according to related legal regulation information, and the nodes in the knowledge graph are traversed in sequence according to the obtained first word segmentation data, so that the nodes of the legal regulations in the knowledge graph are searched according to keywords, and legal regulation contents corresponding to the target nodes with higher relevance are determined to serve as target answers. The nodes can be searched in a traversing way to determine the correlation between the target information and each node, so that the content with higher correlation is obtained and output as the target answer, and the answer rate and the accuracy of the second question-answer module in processing the target information are improved.
Optionally, the third question-answering module is configured to perform correlation processing on the target information to obtain correlation data; extracting the related data based on a machine reading understanding model to obtain candidate answers;
And when the answer probability corresponding to the candidate answer is higher than the answer threshold, the third question-answer module is further used for taking the candidate answer as the target answer.
In the implementation process, when the third question-answering module processes the target information, the target information can be identified according to machine reading, so that a plurality of legal and regulatory information related to the target information can be obtained as related data. And extracting the related data by using a machine reading understanding model to determine candidate answers, selecting according to the answer probability of each candidate answer, and outputting the candidate answer as a target answer when the answer probability is higher than the answer threshold. The method can be used for carrying out information extraction and prediction of response probability based on machine reading understanding, and improves the response rate and accuracy of the third question-answer module when processing the target information.
Optionally, the first question-answering module is connected with the second question-answering module through a first data interface;
the second question-answering module is connected with the third question-answering module through a second data interface;
the third question-answering module is connected with the retrieval module through a third data interface;
the first question-answering module, the second question-answering module, the third question-answering module and the retrieval module are respectively connected with the output interface.
In the implementation process, the question-answering modules can be connected through different data interfaces, and when a target answer cannot be obtained after the previous question-answering module processes according to the target information, the target information is sequentially transmitted to the next question-answering module through the data interfaces to continue to process. The search module is also connected with the third question-answering module through the data interface, so that the third question-answering module does not obtain the target answer, namely, when the serial question-answering module does not obtain the target answer, the target information is continuously transmitted to the search module through the data interface for processing, the target information can be ensured to have the corresponding target answer or search answer, and the answer rate of the target information is improved.
Optionally, the search module is used for performing word segmentation on the target information to obtain second word segmentation data; extracting keyword data based on the second keyword data; and inquiring based on the keyword data to obtain the retrieval answer.
In the implementation process, when the retrieval module processes the target information, in order to improve accuracy in processing, word segmentation processing may be performed on the target information first, so as to extract corresponding keyword data from the obtained second keyword data, so as to query according to the keyword data, and output content with higher relativity content obtained by query as a retrieval answer. When the serial question-answering module can not answer the target information, the relevant content can be searched for answering, so that the use experience of the user is improved.
In a second aspect, an embodiment of the present application further provides an information reply method, where the method is applied to the information reply system described in any one of the foregoing aspects, and the method includes:
processing target information input by an input interface by a serial question-answering module to obtain a target answer;
if the serial question-answering module does not obtain the target answer after processing according to the target information, the serial question-answering module inputs the target information into a retrieval module;
processing according to the target information by the retrieval module to obtain a retrieval answer;
and outputting the target answer or the search answer by an output interface.
In the implementation process, firstly, the input target information is processed through the serial question-answering module, and if the target answer is not obtained, the target information is continuously processed through the retrieval module, and the target answer or the retrieval answer is replied. The target information can be correspondingly processed based on a plurality of different question-answering and retrieval technologies, the answer rate of the target information and the accuracy of the output answer are improved, so that a user can quickly and accurately acquire relevant legal and legal knowledge during learning, the use experience of the user is optimized, and the learning efficiency of the user is improved.
Optionally, the serial question-answering module includes a first question-answering module, a second question-answering module and a third question-answering module;
the serial question-answering module processes the target information input by the input interface to obtain a target answer, and the method comprises the following steps:
the input interface acquires the target information to be replied and inputs the target information into the first question-answering module;
processing the input target information by the first question-answering module;
if the first question-answering module does not obtain the target answer after processing the target information, the first question-answering module inputs the target information into the second question-answering module;
processing the input target information by the second question-answering module;
if the second question-answering module does not obtain the target answer after processing the target information, the second question-answering module inputs the target information into the third question-answering module;
and processing the input target information by the third question-answering module.
In the implementation process, when the serial question-answering modules process the target information, the question-answering modules with three different functions in the serial question-answering modules process the target information in sequence according to a preset sequence, and if the previous question-answering module does not obtain the target answer after processing, the processing is continued in the next question-answering module connected with the target information output until the processing of the third question-answering module is completed. The method can process the target information in turn from the aspects of processing speed, the range of the knowledge base during retrieval, the type of the processed target information and the like, and effectively improves the efficiency and the response rate when generating the target answer.
Optionally, the outputting, by the output interface, the target answer or the search answer includes:
if the first question-answering module, the second question-answering module or the third question-answering module obtains the target answer after being processed based on the target information, outputting the target answer through the output interface;
and if the first question-answering module, the second question-answering module and the third question-answering module do not obtain the target answer after being processed based on the target information, outputting the retrieval answer through the output interface.
In the implementation process, when the output interface outputs the answer, if any question-answering module in the series question-answering module processes the target information to obtain the target answer, the series question-answering module finishes processing, the output interface directly outputs the target answer, and if all question-answering modules in the series question-answering module cannot obtain the target answer after processing the target information, the output interface obtains the search answer of the search module to output, so that the accuracy of information reply is improved.
In a third aspect, embodiments of the present application further provide a computer readable storage medium having stored therein computer program instructions which, when read and executed by a processor, perform the steps of any implementation of the above-described information reply method.
In summary, the embodiments of the present application provide an information reply system, method and computer readable storage medium, which connect a plurality of question-answering modules with different functions in series, so as to sequentially process target information input by a user, and connect a corresponding search module to search information which cannot be answered by the question-answering module, thereby improving the reply rate and the accuracy of the output reply answer when replying to the information.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings that are needed in the embodiments of the present application will be briefly described below, it should be understood that the following drawings only illustrate some embodiments of the present application and should not be considered as limiting the scope, and other related drawings can be obtained according to these drawings without inventive effort for a person skilled in the art.
Fig. 1 is a schematic structural diagram of an information reply system according to an embodiment of the present application;
FIG. 2 is a schematic diagram of another information reply system according to an embodiment of the present application;
fig. 3 is a flow chart of an information reply method according to an embodiment of the present application;
fig. 4 is a detailed flowchart of step S200 according to an embodiment of the present application;
Fig. 5 is a detailed flowchart of step S500 according to an embodiment of the present application.
Icon: 100-an information reply system; 110-an input interface; 120-a serial question-answering module; 130-a retrieval module; 140-output interface; 150-updating the module; 121-a first question-answering module; 122-a second question-answering module; 123-third question-answering module.
Detailed Description
The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. It will be apparent that the described embodiments are only some, but not all, of the embodiments of the present application. All other embodiments, which can be made by one of ordinary skill in the art based on embodiments of the present application without making any inventive effort, are intended to fall within the scope of the embodiments of the present application.
Since related laws and regulations are formulated in each industry, in daily life, many industries and field personnel need to learn and master related laws and regulations to improve their own legal consciousness and perform corresponding actions such as work and study. And when learning the related knowledge of laws and regulations, the efficiency of acquiring the related information in books related to laws and regulations is low. In order to quickly acquire relevant knowledge, a traditional search engine is generally used at present, a mode of inputting keywords to perform search is adopted to acquire corresponding search results, but due to the fact that the existing laws and regulations in China are numerous and have the characteristics of complexity and the like, the traditional search engine is difficult to accurately and quickly acquire laws and regulations information which is higher in correlation with the keyword information input by a user, the retrieval capability of related problems aiming at the laws and regulations information is poor, the problems raised by the user cannot be accurately and quickly solved, experience of the user in use is reduced, and the efficiency of learning the laws and regulations knowledge by the user is affected.
In order to solve the above problems, the embodiment of the present application provides an information reply system, where the information reply system may be disposed in an electronic device, and the electronic device may be an electronic device with a logic computing function, such as a server, a personal computer (Personal Computer, PC), a tablet computer, a smart phone, a personal digital assistant (Personal Digital Assistant, PDA), etc., and may obtain information of a related problem input by a user, process the information, and then quickly and accurately output a reply answer for the user to refer to and learn.
Alternatively, an input-output unit may be provided in the electronic device for enabling a user to input data, such as a mouse, a keyboard, and the like. The electronic device may also be provided with a display unit to provide an interactive interface (e.g., a user operation interface) between the electronic device and the user, and to display answers to the user's references. The display unit may be a liquid crystal display or a touch display, for example. In the case of a touch display, the touch display may be a capacitive touch screen or a resistive touch screen, etc. supporting single-point and multi-point touch operations. Supporting single-point and multi-point touch operations means that the touch display can sense touch operations simultaneously generated from one or more positions on the touch display, and the sensed touch operations are passed to the processor for calculation and processing.
Referring to fig. 1, fig. 1 is a schematic structural diagram of an information reply system according to an embodiment of the present application, and an information reply system 100 may include: an input interface 110, a serial question-and-answer module 120, a retrieval module 130, and an output interface 140; the input interface 110 is connected with the serial question-answering module 120, the output interface 140 is connected with the serial question-answering module 120 and the retrieval module 130, and the serial question-answering module 120 is connected with the retrieval module 130.
Wherein the input interface 110 is used for receiving target information to be replied to. The input interface 110 may obtain target information input by a user through an input/output unit in the electronic device, where the target information may be various types of data, such as text, code, and voice, and may be various forms, such as sentences, keywords, and the like. After the input interface 110 obtains the target information, the target information can be transmitted to the serial question-and-answer module 120 through the data interface for relevant processing.
The serial question-answering module 120 may include a first question-answering module 121, a second question-answering module 122, and a third question-answering module 123 of different functions, which are serially connected according to a preset order, and the serial question-answering module 120 is configured to process target information to obtain a target answer.
It should be noted that, the first question-answering module 121 may include a common question set question-answering module, the second question-answering module 122 may include a knowledge-graph question-answering module, and the third question-answering module 123 may include a machine-readable understanding question-answering module.
Alternatively, the preset order may be an order set according to the functional characteristics of the three question-answering modules, and in order to further improve the efficiency and the answer rate when the series question-answering module 120 generates the target answer, the preset order of the series of the three question-answering modules with different functions may be defined. For example, the first question-answering module 121 for processing the target information with the most rapid processing speed is used as the first question-answering module for processing the target information, the second question-answering module 122 for processing the target information with the second question-answering module for retrieving answers from the knowledge base of the knowledge base is used as the third question-answering module 123 for processing the target information with the last machine-reading understanding question-answering module for retrieving and extracting a plurality of unstructured text data to obtain answers, so that the target information can be processed sequentially from multiple aspects such as the processing speed, the range of the knowledge base during retrieving, the type of the processed target information, and the like, thereby effectively improving the efficiency and the answer rate when generating the target answers.
Optionally, the common question set question-answering module is a question-answering module based on a common question set (Frequently Asked Questions, FAQ) capable of organizing common questions into question-answering pairs to construct a corresponding question-answering database to answer information. The knowledge graph question-answering module is a question-answering module based on knowledge graph questions-answering (Knowledge Graph Question Answering, KBQA) and can directly retrieve answers from the constructed structured knowledge graph. The machine-readable understanding question-answering module is a text-based question-answering module, also known as a machine-readable understanding (Machine Reading Comprehension, MRC) question-answering module, each question corresponding to a plurality of unstructured text data from which answers are retrieved and extracted.
When the target answer is not obtained after the serial question-answering module 120 processes the target information, the retrieving module 130 is configured to process the target information to obtain a retrieving answer. The processing of the target information by the connected search module 130 can be continued when the serial question-answering module 120 cannot answer the target information, so as to obtain a corresponding search answer for replying.
The output interface 140 is configured to output a target answer or a search answer, and can output the target answer or the search answer in real time when the serial question-answer module 120 obtains the target answer or the search answer is obtained by the search module 130, where the output target answer or the search answer can be output in various forms, for example, can be displayed in text form in a display unit of the electronic device for a user to view the target answer or the search answer, or can be played in voice form in a play unit of the electronic device for the user to hear the target answer or the search answer.
Through the serial connection and the connection of the question-answering modules with different functions and the retrieval module, the target information can be correspondingly processed based on a plurality of different question-answering and retrieval technologies, the answer rate of the target information and the accuracy of the output answer are improved, so that a user can quickly and accurately acquire relevant legal and legal knowledge during learning, the use experience of the user is optimized, and the learning efficiency of the user is improved.
Alternatively, when the first question-answering module 121 processes the target information, since there may be an abnormal situation of the target information input by the user, for example, the target information is an empty string, etc., the first question-answering module 121 may first determine whether the target information is normal; if the target information is abnormal, for example, if the target information is an empty string, the first question-answering module 121 may be configured to feed back a re-input request, so as to prompt the user to re-input the information through the re-input request, thereby reducing the abnormal situation caused by abnormal input.
If the target information is normal, the first question-answer module 121 is further configured to determine whether the preset question-answer database has standard information with similarity to the target information higher than a similarity threshold based on the classification model.
It should be noted that, before the first question-answering module 121 processes the target information, a corresponding question-answering database may also be constructed according to the actual content of law and regulation. Question-answer pairs of input information and associated output answers to frequently retrieved questions may be stored in the question-answer database. The method can construct a corresponding information template by combining manual labeling and rule matching, and acquire answers from legal and legal texts so as to improve the establishment efficiency of a question-answer database.
By way of example, the specific steps of building a question-answer database may include: firstly, abstracting an information template according to a known problem; then, converting the terms in the law and regulation into different question forms to form standard question-answer pairs; finally, the constructed information templates and standard question-answer pairs are stored in a question-answer database for retrieval. Common information templates may include: what is something; something, etc. When the question-answer database is established, a plurality of collected laws and regulations of each industry can be correspondingly analyzed and processed, so that a plurality of information templates and corresponding standard question-answer pairs are constructed. In addition, the standard question-answer pairs in the question-answer database can be updated continuously in consideration of the fact that a plurality of information templates and corresponding standard question-answer pairs in the question-answer database are difficult to cover more question types.
When the first question-answering module 121 processes the target information and a plurality of information templates in the question-answering database, the target information may be analyzed using the classification model to find and match standard information similar to the target information. For example, the classification model may be a BERT model, and when the classification model processes the target information, the training data set may be constructed first, that is, similar question sentences and dissimilar question sentences may be constructed for each question sentence in the information templates and standard answers that are capable of being in the question-answer database. Wherein the dissimilar question is derived from the dataset itself, i.e. for a question, a question different from the question is selected from the dataset as its dissimilar question. For the construction of similar questions, the SimBERT model may be used for generation. The SimBERT model is a model which is obtained by integrating search and generation based on a UniLM idea on the basis of the BERT model and can be further finely tuned, and has similar question generation capability. Specific steps of data set construction may include: dividing a plurality of information templates into a training set, a verification set and a test set according to a preset proportion, for example, a proportion of 8:1:1 by using a question sentence generated by the information templates; after the legal and legal contents are subjected to word segmentation, the word vector of each word is obtained as training expectation of a word2vec model (training is carried out by selecting a CBOW model) so as to obtain the semantic vector of each question sentence; in each data set, calculating the similarity between the question sentence and the rest of question sentences by using cosine similarity for each question sentence, selecting the first 5 sentences with high similarity as dissimilar question sentences of the question sentence, and marking the dissimilar question sentences as 0; for each question, generating 5 similar questions by using a SimBERT model, and marking the similar questions as 1; after the data set construction is completed, the BERT model is trained using a training set, the verification set is used to tune the parameters to obtain the best model, and the test set is used to evaluate the performance of the model. Therefore, the similarity problem and the dissimilarity problem of the target information and the plurality of information templates can be divided by the classification model, and the corresponding similarity threshold value is set, and the information templates exceeding the similarity threshold value are used as standard information.
If the question-answer database has standard information, the first question-answer module 121 is further configured to use the standard answer of the standard information as the target answer. Since the standard information has high similarity with the target information, the standard answer corresponding to the standard information can be used as the target answer for replying to the target information and output. The target information can be quickly and accurately matched with whether the corresponding target answer exists or not through accurate processing of the classification model, and efficiency and accuracy of the first question-answering module in answering are effectively improved.
Optionally, referring to fig. 2, fig. 2 is a schematic structural diagram of another information reply system provided in an embodiment of the present application, where the information reply system may further include: the update module 150, the update module 150 may be connected with the first question-answering module. The output target answers and the search answers can be stored in a background database for the staff to analyze, and the updating module 150 can detect the target answers or the search answers corresponding to the target information according to the analysis results of the staff to judge whether the target answers or the search answers are correct answers of the target information.
If the target answer or the search answer corresponding to the target information is wrong, the updating module 150 is specifically configured to obtain a correct answer, and add the target information and the correct answer to the question-answer database of the first question-answer module 121. The updating module 150 can obtain a correct answer corresponding to the target information from the analysis result of the staff, and add the target information and the associated correct answer to the question-answer database of the first question-answer module 121, and can collect and store the information and the answer of the wrong answer, so that the first question-answer module 121 is continuously and iteratively updated, the question-answer database of the first question-answer module is continuously perfected, and the answer rate and the accuracy of the target information when being processed in the first question-answer module 121 are further improved.
Alternatively, when the second question-answering module 122 processes the target information, in order to improve accuracy in processing, word segmentation may be performed on the target information first to obtain the first word segmentation data. The second question-answering module 122 may perform word segmentation based on various word segmentation tools, for example, using a Jieba word segmentation tool to segment a user question, and remove stop words of target information to obtain first word segmentation data.
It should be noted that, since the second question-answering module 122 is a knowledge-graph question-answering module, relevant nodes can be found from the constructed knowledge graph according to the information of the questions posed by the user, and the node content is returned to the user as an answer. Because the existing laws and regulations are numerous, it is difficult to construct a general ontology suitable for all laws and regulations, therefore, a catalog map can be constructed for each laws and regulations according to the thought of information retrieval, and a plurality of catalog maps form a complete preset knowledge map.
Optionally, the concept of the catalog map includes laws, chapters, clauses, knowledge points. The catalog map can take the chapter names of various laws and regulations as nodes, and the clauses in the chapter are taken as the next-level nodes. If the clause has the next level content, the clause content is regarded as a next level node, and the part other than the clause content is regarded as a current level node. For each item of clause content, the clause content can be split into a plurality of clauses, knowledge points in the clauses are extracted as attributes by combining possible problem information, and a catalog map is constructed by a < clause, what and knowledge point > triplet.
The second question-answering module 122 is further configured to determine whether a target node, of the plurality of nodes, having a relevance to the first word segmentation data higher than a relevance threshold, is included in traversing among the plurality of nodes of the preset knowledge graph based on the first word segmentation data.
Alternatively, cosine similarity may be used for calculation in order to find nodes related to the target information in the knowledge-graph. Since the nodes in the constructed directory map store a sentence or text fragment, not a named entity. Therefore, in the process of searching the nodes according to the user problems, named entity identification is not needed for the user problems, the relevance between the user problems after word segmentation and removal of the stop words and the node content is only needed to be calculated, the node with low relevance is removed according to a preset relevance threshold value, the node with highest relevance is selected as a target node, and the content in the target node is used as a target answer. The method for calculating the correlation between the target information and the contents of each node by using the cosine similarity may include: suppose that the user problem is q (x 1 ,x 2 ,…,x m ) One node content is d (x' 1 ,x′ 2 ,…,x′ n ) Asking users using a bag of words modelThe question and node contents are represented as one-dimensional vectors q (v 1 ,v 2 ,…,v o ) And d (v' 1 ,v′ 2 ,…,v′ o ) The cosine similarity calculation formula between the user problem and the node content is:
wherein, |q (v ′ 1 ,v ′ 2 ,…,v ′ o ) I and d (v) ′ 1 ,v ′ 2 ,…,v ′ o ) I represent vector q (v) 1 ,v 2 ,…,v o ) And d (v' 1 ,v′ 2 ,…,v′ o ) Is a die length of the die. If the plurality of nodes have target nodes, the second question-answer module 122 is further configured to select a node answer corresponding to the target node with the highest relevance as the target answer. The nodes can be searched in a traversing way to determine the correlation between the target information and each node, so that the content with higher correlation is obtained and output as the target answer, and the answer rate and the accuracy of the second question-answer module in processing the target information are improved.
Alternatively, when processing the target information, the third question-answering module 123 may perform correlation processing on the target information to obtain the relevant data. Keywords in the target information can be extracted, various terms in laws and regulations are searched according to the keywords, and then the terms of laws and regulations with higher relevance are screened out to serve as relevant data for extracting candidate answers according to the relevance ranking.
It should be noted that, since the third question-answering module 123 is a machine-readable understanding question-answering module, several relevant terms in law and regulation can be retrieved from the search engine according to the target information of the question posed by the user, then an answer segment is extracted from each term using the machine-readable understanding model, and the optimal answer is selected as the final answer, and returned to the user. Therefore, a corresponding machine reading understanding model needs to be built in advance, and in view of the excellent performance of the pre-training language model in the machine reading understanding task, the machine reading understanding model is built by using the RoBERTa pre-training language model, so that the third question-answering module 123 can extract relevant data based on the machine reading understanding model to obtain candidate answers.
Alternatively, the pre-trained language model may take each legal and regulatory term as a Document (Document), splice together target information corresponding to the user Question (Question) and the Document, and add segmentation symbols [ SEP ] in the middle and at the end]And adding special classification symbol [ CLS ] before target information]. Assuming that the length of the target information is M and the length of Document is N, transmitting the spliced text sequence into a pre-training language model to obtain semantic vector representation H of each Token in the text sequence i (i＝[CLS],1,2,…,m,[SEP],1,2,…,n,[SEP]). The hidden layer vector corresponding to Document is expressed as H i (i=1, 2,3, …, n) respectively by two linear layers and SoftMax calculation to obtain probability S of each Token as answer start and end position i (i=1, 2,3, …, n) and E j (j=1, 2,3, …, n) and then find the S i *E j (i,j＝1,2,…,n,i<j) The largest combination takes the text segment between positions i and j in Document as a candidate answer. S is S i *E j Is a predictive score (HA score) of the candidate answer. In order to judge whether the Document contains the answer, the answer probability corresponding to the candidate answer can be calculated, the answer probability can comprise answer probability and no answer probability, and the method is [ CLS ]]Vector representation H of position [CLS] After a linear layer and SoftMax calculation, no answer probability (NA Score) is obtained, and whether answer is required to be refused is judged according to the no answer probability and a manually set threshold.
Predictive value S of start and end position during training i (i=1, 2,3, …, n) and E j (j=1, 2,3, …, n) and the true value Y start And Y end Between which two cross entropy loss values loss will be calculated start And loss of end Judging whether the Document contains the predicted value P of the answer NA And true value Y NA Will also get a cross entropy loss d loss value loss NA The calculation mode may include:
loss start ＝CE(S i ,Y start )(i＝1,2,3,…,n)；
loss end ＝CE(E j ,Y end )(j＝1,2,3,…,n)；
loss NA ＝CE(P NA ,Y NA )；
wherein the final loss value is loss start 、loss end And loss of NA The average value of the three can be calculated by the following steps:
for example, when the data set used contains 622 pieces of data in total, it may be divided into a training set, a validation set, and a test set in a ratio of 8:1:1. So that the pre-training language model learns on 497 pieces of training data and carries out parameter adjustment on 62 pieces of verification data to obtain the machine reading understanding model with better effect.
Alternatively, whether to reject the answer may be determined according to the answer-free probability and the threshold, and the third question-answer module 123 may be further configured to target the candidate answer when the answer probability corresponding to the candidate answer is higher than the answer threshold. And the answers with highest predictive scores can be selected as target answers, so that the situation of wrong answers is reduced. The method can be used for carrying out information extraction and prediction of response probability based on machine reading understanding, and improves the response rate and accuracy of the third question-answer module when processing the target information.
Alternatively, the first question-answering module 121 and the second question-answering module 122 may be connected through a first data interface; the second question-answering module 122 and the third question-answering module 123 may be connected through a second data interface; the third question and answer module 123 and the retrieval module 130 may be connected through a third data interface; the first question-answering module 121, the second question-answering module 122, the third question-answering module 123, and the retrieval module 130 are connected to the output interface 140, respectively.
The question-answering modules can be connected through different data interfaces, so that when a target answer cannot be obtained after the previous question-answering module processes according to the target information, the target information is sequentially transmitted to the next question-answering module through the data interfaces to continue processing. The search module is also connected with the third question-answering module through the data interface, so that the third question-answering module does not obtain the target answer, namely, when the serial question-answering module does not obtain the target answer, the target information is continuously transmitted to the search module through the data interface for processing, the target information can be ensured to have the corresponding target answer or search answer, and the answer rate of the target information is improved.
Optionally, when the retrieval module 130 processes the target information, in order to improve accuracy during processing, word segmentation may be performed on the target information to obtain second word segmentation data. The jeba word segmentation tool can be used for word segmentation of the target information, and stop words in the target information are removed, so that corresponding second word segmentation data are obtained.
The retrieval module 130 is further configured to extract keyword data based on the second keyword data, and may use a variety of keyword extraction algorithms, for example, textRank algorithm to extract the keyword data in the second keyword data.
The search module 130 is further configured to query based on the keyword data to obtain a search answer. The extracted keyword data may be used to query in a search engine, for example, an ES search engine may construct an ES query sentence to find relevant documents in the ES search engine, and return one or more results with highest relevance as search answers. Before searching, the legal and legal documents can be segmented in a rule matching mode, then the segmented paragraphs are segmented, keywords in the paragraphs are extracted, and finally the paragraph contents and the keywords are stored in a database together for subsequent searching.
The embodiment of the application also provides an information reply method, which is applied to any one of the information reply systems provided by the application, referring to fig. 3, fig. 3 is a flow diagram of the information reply method provided by the embodiment of the application, and the method can include: steps S200-S500.
Step S200, the target information input by the input interface is processed by the serial question-answering module to obtain a target answer.
And step S300, if the serial question-answering module does not obtain the target answer after processing according to the target information, the serial question-answering module inputs the target information into the retrieval module.
Step S400, processing is carried out by the retrieval module according to the target information to obtain a retrieval answer.
Step S500, the output interface outputs the target answer or the search answer.
The method comprises the steps of firstly processing input target information through a serial question-answer module in an information answer system, and if a target answer is not obtained, continuing processing the target information through a retrieval module, and replying with the target answer or the retrieval answer. The target information can be correspondingly processed based on a plurality of different question-answering and retrieval technologies, the answer rate of the target information and the accuracy of the output answer are improved, so that a user can quickly and accurately acquire relevant legal and legal knowledge during learning, the use experience of the user is optimized, and the learning efficiency of the user is improved.
Optionally, the serial question-answering module may include a first question-answering module, a second question-answering module, and a third question-answering module; referring to fig. 4, fig. 4 is a detailed flowchart of step S200 according to an embodiment of the present application, and step S200 may include steps S210-S260.
In step S210, the input interface acquires the target information to be replied to, and inputs the target information into the first question-answering module.
Step S220, the first question-answering module processes the input target information.
In step S230, if the first question-answering module does not obtain the target answer after processing the target information, the first question-answering module inputs the target information into the second question-answering module.
Step S240, the second question-answering module processes the input target information.
In step S250, if the second question-answering module does not obtain the target answer after processing the target information, the second question-answering module inputs the target information into the third question-answering module.
In step S260, the third question-answering module processes the input target information.
When the serial question-answering module processes the target information, the question-answering modules with three different functions in the serial question-answering module can process the target information in sequence according to a preset sequence. And when the target answer is not obtained after the processing of the first question-answering module, inputting the target information into the second question-answering module for continuous processing, and when the target answer is not obtained after the processing of the second question-answering module, inputting the target information into the third question-answering module for continuous processing. The method can process the target information in turn from the aspects of processing speed, the range of the knowledge base during retrieval, the type of the processed target information and the like, and effectively improves the efficiency and the response rate when generating the target answer.
Optionally, referring to fig. 5, fig. 5 is a detailed flowchart of step S500 provided in an embodiment of the present application, and step S500 may include steps S510-S520.
Step S510, if the first question-answering module, the second question-answering module or the third question-answering module obtains the target answer after processing based on the target information, the output interface outputs the target answer.
In step S520, if the first question-answering module, the second question-answering module, and the third question-answering module do not obtain the target answer after processing based on the target information, the output interface outputs the search answer.
When the output interface outputs the answers, if any one of the question-answering modules in the series connection processes the target information to obtain the target answer, the question-answering module in the series connection ends the processing, the output interface directly outputs the obtained target answer, and if all the question-answering modules in the series connection cannot obtain the target answer after processing the target information, the output interface obtains the retrieval answer of the retrieval module to output, so that the accuracy of information reply is improved.
The embodiment of the application also provides a computer readable storage medium, wherein the computer readable storage medium stores computer program instructions, and when the computer program instructions are read and executed by a processor, the steps in any one of the information reply methods provided in the embodiment are executed.
In the several embodiments provided in the present application, it should be understood that the disclosed apparatus may be implemented in other manners. The apparatus embodiments described above are merely illustrative, for example, block diagrams in the figures illustrate the architecture, functionality, and operation of possible implementations of devices according to various embodiments of the present application. In this regard, each block in the block diagrams may represent a module, segment, or portion of code, which comprises one or more executable instructions for implementing the specified logical function(s). It should also be noted that in some alternative implementations, the functions noted in the block may occur out of the order noted in the figures. For example, two blocks shown in succession may, in fact, be executed substantially concurrently, or the blocks may sometimes be executed in the reverse order, depending upon the functionality involved. It will also be noted that each block of the block diagrams, and combinations of blocks in the block diagrams, can be implemented by special purpose hardware-based systems that perform the specified functions or acts, or combinations of special purpose hardware and computer instructions.
In addition, functional modules in the embodiments of the present application may be integrated together to form a single part, or each module may exist alone, or two or more modules may be integrated to form a single part.
The functions, if implemented in the form of software functional modules and sold or used as a stand-alone product, may be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application may be embodied essentially or in a part contributing to the prior art or in a part of the technical solution, in the form of a software product stored in a storage medium, comprising several instructions for causing a computer device (which may be a personal computer, a server, a network device, etc.) to perform all or part of the steps of the method according to the embodiments of the present application. And the aforementioned storage medium includes: a U-disk, a removable hard disk, a Read-Only Memory (ROM), a random access Memory (RAM, random Access Memory), a magnetic disk, or an optical disk, or other various media capable of storing program codes.
The above description is only an example of the present application and is not intended to limit the scope of the present application, and various modifications and variations will be apparent to those skilled in the art. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included in the protection scope of the present application. It should be noted that: like reference numerals and letters denote like items in the following figures, and thus once an item is defined in one figure, no further definition or explanation thereof is necessary in the following figures.
The foregoing is merely illustrative of the present application, and the present application is not limited thereto, and any person skilled in the art will readily recognize that variations or substitutions are within the scope of the present application.
It is noted that relational terms such as first and second, and the like are used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Moreover, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising … …" does not exclude the presence of other like elements in a process, method, article or apparatus that comprises the element.
Claims (12)
1. An information reply system, the system comprising: the system comprises an input interface, a serial question-answering module, a retrieval module and an output interface;
the input interface is connected with the serial question-answering module, the output interface is connected with the serial question-answering module and the retrieval module, and the serial question-answering module is connected with the retrieval module;
the input interface is used for receiving target information to be replied;
the serial question-answering module comprises a first question-answering module, a second question-answering module and a third question-answering module which are connected in series according to a preset sequence, wherein the serial question-answering module is used for processing the target information to obtain a target answer;
the retrieval module is used for processing the target information to obtain a retrieval answer when the target information is not obtained after the serial question-answering module processes the target information;
the output interface is used for outputting the target answer or the search answer.
2. The system of claim 1, wherein the first question-answering module comprises a common question set question-answering module, the second question-answering module comprises a knowledge-graph question-answering module, and the third question-answering module comprises a machine-readable understanding question-answering module.
3. The system of claim 2, wherein the first question-answering module is configured to determine whether the target information is normal;
if the target information is abnormal, the first question-answering module is further used for feeding back a re-input request;
if the target information is normal, the first question-answer module is further used for determining whether standard information with similarity higher than a similarity threshold value with the target information exists in a preset question-answer database or not based on a classification model; wherein, the question-answer database comprises a plurality of information templates;
and if the standard information is in the question-answer database, the first question-answer module is further used for taking the standard answer of the standard information as the target answer.
4. A system according to claim 3, wherein the system further comprises: the updating module is used for detecting the target answer or the search answer corresponding to the target information;
if the target answer or the search answer corresponding to the target information is wrong, the updating module is specifically configured to obtain a correct answer, and add the target information and the correct answer to the question-answer database of the first question-answer module.
5. The system of claim 2, wherein the second question-answering module is configured to perform word segmentation on the target information to obtain first word segmentation data; traversing a plurality of nodes of a preset knowledge graph based on the first word segmentation data, and determining whether target nodes with correlation with the first word segmentation data higher than a correlation threshold value exist in the plurality of nodes;
and if the target node exists in the plurality of nodes, the second question-answering module is further used for taking the node answer corresponding to the target node as the target answer.
6. The system of claim 2, wherein the third question-answering module is configured to perform correlation processing on the target information to obtain correlation data; extracting the related data based on a machine reading understanding model to obtain candidate answers;
and when the answer probability corresponding to the candidate answer is higher than the answer threshold, the third question-answer module is further used for taking the candidate answer as the target answer.
7. The system of any one of claims 1-6, wherein the first question-answering module is connected with the second question-answering module through a first data interface;
The second question-answering module is connected with the third question-answering module through a second data interface;
the third question-answering module is connected with the retrieval module through a third data interface;
the first question-answering module, the second question-answering module, the third question-answering module and the retrieval module are respectively connected with the output interface.
8. The system of claim 1, wherein the retrieval module is configured to perform word segmentation on the target information to obtain second word segmentation data; extracting keyword data based on the second keyword data; and inquiring based on the keyword data to obtain the retrieval answer.
9. An information reply method, characterized in that the method is applied to the information reply system according to any one of claims 1 to 8, the method comprising:
processing target information input by an input interface by a serial question-answering module to obtain a target answer;
if the serial question-answering module does not obtain the target answer after processing according to the target information, the serial question-answering module inputs the target information into a retrieval module;
processing according to the target information by the retrieval module to obtain a retrieval answer;
And outputting the target answer or the search answer by an output interface.
10. The method of claim 9, wherein the series of question-answering modules includes a first question-answering module, a second question-answering module, and a third question-answering module;
the serial question-answering module processes the target information input by the input interface to obtain a target answer, and the method comprises the following steps:
the input interface acquires the target information to be replied and inputs the target information into the first question-answering module;
processing the input target information by the first question-answering module;
if the first question-answering module does not obtain the target answer after processing the target information, the first question-answering module inputs the target information into the second question-answering module;
processing the input target information by the second question-answering module;
if the second question-answering module does not obtain the target answer after processing the target information, the second question-answering module inputs the target information into the third question-answering module;
and processing the input target information by the third question-answering module.
11. The method of claim 9, wherein outputting, by the output interface, the target answer or the search answer comprises:
if the first question-answering module, the second question-answering module or the third question-answering module obtains the target answer after being processed based on the target information, outputting the target answer through the output interface;
and if the first question-answering module, the second question-answering module and the third question-answering module do not obtain the target answer after being processed based on the target information, outputting the retrieval answer through the output interface.
12. A computer readable storage medium, characterized in that the computer program instructions are stored in the readable storage medium, which computer program instructions, when being executed by a processor, perform the steps of the method according to any of claims 9-11.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310147714.4A CN116701579A (en) | 2023-02-21 | 2023-02-21 | Information reply system, method and computer readable storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310147714.4A CN116701579A (en) | 2023-02-21 | 2023-02-21 | Information reply system, method and computer readable storage medium |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116701579A true CN116701579A (en) | 2023-09-05 |
Family
ID=87839870
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310147714.4A Pending CN116701579A (en) | 2023-02-21 | 2023-02-21 | Information reply system, method and computer readable storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116701579A (en) |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108717433A (en) * | 2018-05-14 | 2018-10-30 | 南京邮电大学 | A kind of construction of knowledge base method and device of programming-oriented field question answering system |
| CN109885672A (en) * | 2019-03-04 | 2019-06-14 | 中国科学院软件研究所 | A kind of question and answer mode intelligent retrieval system and method towards online education |
| CN109885660A (en) * | 2019-02-22 | 2019-06-14 | 上海乐言信息科技有限公司 | A kind of question answering system and method based on information retrieval that knowledge mapping is energized |
| CN111737443A (en) * | 2020-08-14 | 2020-10-02 | 支付宝(杭州)信息技术有限公司 | Answer text processing method and device and key text determining method |
| CN112035638A (en) * | 2020-08-28 | 2020-12-04 | 腾讯科技（深圳）有限公司 | Information processing method, device, storage medium and equipment |
| CN112597291A (en) * | 2020-12-26 | 2021-04-02 | 中国农业银行股份有限公司 | Intelligent question and answer implementation method, device and equipment |
| CN115238101A (en) * | 2022-09-23 | 2022-10-25 | 中国电子科技集团公司第十研究所 | Multi-engine intelligent question-answering system oriented to multi-type knowledge base |
-
2023
- 2023-02-21 CN CN202310147714.4A patent/CN116701579A/en active Pending
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108717433A (en) * | 2018-05-14 | 2018-10-30 | 南京邮电大学 | A kind of construction of knowledge base method and device of programming-oriented field question answering system |
| CN109885660A (en) * | 2019-02-22 | 2019-06-14 | 上海乐言信息科技有限公司 | A kind of question answering system and method based on information retrieval that knowledge mapping is energized |
| CN109885672A (en) * | 2019-03-04 | 2019-06-14 | 中国科学院软件研究所 | A kind of question and answer mode intelligent retrieval system and method towards online education |
| CN111737443A (en) * | 2020-08-14 | 2020-10-02 | 支付宝(杭州)信息技术有限公司 | Answer text processing method and device and key text determining method |
| CN112035638A (en) * | 2020-08-28 | 2020-12-04 | 腾讯科技（深圳）有限公司 | Information processing method, device, storage medium and equipment |
| CN112597291A (en) * | 2020-12-26 | 2021-04-02 | 中国农业银行股份有限公司 | Intelligent question and answer implementation method, device and equipment |
| CN115238101A (en) * | 2022-09-23 | 2022-10-25 | 中国电子科技集团公司第十研究所 | Multi-engine intelligent question-answering system oriented to multi-type knowledge base |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110502621B (en) | Question answering method, question answering device, computer equipment and storage medium | |
| CN111475623B (en) | Case Information Semantic Retrieval Method and Device Based on Knowledge Graph | |
| US9613166B2 (en) | Search suggestions of related entities based on co-occurrence and/or fuzzy-score matching | |
| CN112667794A (en) | Intelligent question-answer matching method and system based on twin network BERT model | |
| CN111639171A (en) | Knowledge graph question-answering method and device | |
| US20120078918A1 (en) | Information Relation Generation | |
| CN112035730B (en) | Semantic retrieval method and device and electronic equipment | |
| CN112800170A (en) | Question matching method and device and question reply method and device | |
| CN111368048A (en) | Information acquisition method and device, electronic equipment and computer readable storage medium | |
| CN112328800A (en) | System and method for automatically generating programming specification question answers | |
| CN111125295B (en) | LSTM-based method and system for obtaining answers to food safety questions | |
| CN111813930B (en) | Similar document retrieval method and device | |
| CN112270188A (en) | Questioning type analysis path recommendation method, system and storage medium | |
| CN115204156A (en) | Keyword extraction method and device | |
| CN113868406B (en) | Search method, search system, and computer-readable storage medium | |
| US9507834B2 (en) | Search suggestions using fuzzy-score matching and entity co-occurrence | |
| CN110750632B (en) | Improved Chinese ALICE intelligent question-answering method and system | |
| CN116662518A (en) | Question answering method, question answering device, electronic equipment and readable storage medium | |
| CN116562280A (en) | Literature analysis system and method based on general information extraction | |
| CN116049376A (en) | Method, device and system for retrieving and replying information and creating knowledge | |
| Hall | Opportunities and risks in digital humanities research | |
| CN116701579A (en) | Information reply system, method and computer readable storage medium | |
| CN116414940A (en) | Standard problem determining method and device and related equipment | |
| Bulfamante | Generative enterprise search with extensible knowledge base using AI | |
| CN111159366A (en) | Question-answer optimization method based on orthogonal theme representation |
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 |