CN111368042A - Intelligent question and answer method and device, computer equipment and computer storage medium - Google Patents
Intelligent question and answer method and device, computer equipment and computer storage medium Download PDFInfo
- Publication number
- CN111368042A CN111368042A CN202010091180.4A CN202010091180A CN111368042A CN 111368042 A CN111368042 A CN 111368042A CN 202010091180 A CN202010091180 A CN 202010091180A CN 111368042 A CN111368042 A CN 111368042A
- Authority
- CN
- China
- Prior art keywords
- document
- question
- word
- sentence
- answer
- 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
Images
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
- G06F16/31—Indexing; Data structures therefor; Storage structures
- G06F16/316—Indexing structures
- G06F16/319—Inverted lists
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Databases & Information Systems (AREA)
- General Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Artificial Intelligence (AREA)
- Human Computer Interaction (AREA)
- Computational Linguistics (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
- Machine Translation (AREA)
Abstract
The application discloses an intelligent question-answering method, an intelligent question-answering device, computer equipment and a computer storage medium, relates to the technical field of artificial intelligence, and can understand question sentences input by a user, analyze retrieval intention of the user and provide high-quality answer sentences for the user. The method comprises the following steps: when a question statement is received, acquiring a related document with the matching degree of the question statement ranked before a preset numerical value from a pre-arranged knowledge base; forming an input sentence by each associated document and the question sentence, inputting the input sentence into a pre-trained reading understanding model, and predicting the probability value of each part intercepted from the associated document as an answer sentence; and generating the output answer sentence by using each part intercepted from the associated document as the probability value of the answer sentence.
Description
Technical Field
The present invention relates to the field of artificial intelligence technology, and in particular, to an intelligent question answering method, apparatus, computer device, and computer storage medium.
Background
With the rapid development of the internet, the conventional search engine based on the keywords cannot well meet the internet information retrieval requirements due to the defects of low accuracy, redundant information, the need of screening search results by users and the like. The question-answering system allows a user to use natural language question to inquire information, can understand the question input by the user, analyzes the search intention of the user, provides high-quality answers, not only accords with the search habits of people, but also improves the efficiency of information inquiry.
The open-domain intelligent question answering is a form of a question answering system, and mainly adopts the similarity matching of the questions and answers to linguistic data, or knowledge graph retrieval and deep learning-based generation type question answering, so that the question answering of basic sentences can be realized. However, in the face of the open domain questions and answers of the day, month and day, and various kinds, based on the similarity matching of the questions and answers to the corpus, or based on the knowledge map retrieval, a large-scale question and answer corpus, or a knowledge map triple corpus needs to be maintained, so that the knowledge base often has problems that the coverage is incomplete, the updating is not timely, the user cannot be answered, the accuracy of the generated questions and answers cannot meet the requirement of smooth conversation, and the accurate answers cannot be given quickly.
Disclosure of Invention
In view of the above, the present invention provides an intelligent question answering method, an intelligent question answering device, a computer device and a computer storage medium, and mainly aims to solve the problem that the accuracy of the current generated question answering cannot achieve smooth conversation.
According to an aspect of the present invention, there is provided an intelligent question answering method, including:
when a question statement is received, acquiring a related document with the matching degree of the question statement ranked before a preset numerical value from a pre-arranged knowledge base;
combining each part intercepted from each associated document and question sentences into an input sentence, inputting the input sentence into a pre-trained reading understanding model, and predicting the probability value of each part intercepted from each associated document as an answer sentence;
and generating the output answer sentence by using each part intercepted from the associated document as the probability value of the answer sentence.
Further, the document sets sorted from the websites are recorded in the pre-sorted knowledge base, and before the associated documents with the matching degree ranking before the preset value with the question sentences are obtained from the pre-sorted knowledge base when the question sentences are received, the method further comprises the following steps:
performing word segmentation processing on the document set in the pre-sorted knowledge base, and establishing an inverted index from word segmentation to each document in the document set;
and performing word frequency statistics on the participles of each document in the document set by using a word bag model to obtain the word frequency of the participles in each document.
Further, the obtaining of the associated document ranked before the preset numerical value with the matching degree of the question statement from the pre-sorted knowledge base specifically includes:
calculating evaluation values of importance degrees of the problem sentences in all the documents based on the inverted indexes of the documents in the document set from the word segmentation;
and sorting the evaluation values from large to small, and selecting the document with the evaluation value ranking before a preset numerical value as an associated document.
Further, the calculating an evaluation value of the importance degree of the question statement in each document based on the inverted index of the word segmentation to each document in the document set specifically includes:
performing word segmentation processing on the question sentences, and inquiring word frequency of each word in the question sentences in each document in the document set and word segmentation amount contained in each document based on the inverted indexes of each document in the document set from the word segmentation;
calculating the evaluation value of the importance degree of each participle in the question sentence in each document according to the number of documents in the document set, the word frequency of each participle in the question sentence in each document in the document set and the word quantity contained in each document;
and summarizing the evaluation value of the importance degree of each participle in the question sentence in each document to obtain the evaluation value of the importance degree of the question sentence in each document.
Further, the pre-trained reading understanding model performs fine-tune training and prediction of a reading understanding task on the question-answer data set by using a bert pre-training model, and includes a pre-training stage and a reading understanding stage, wherein each part intercepted from each associated document and a question sentence form an input sentence, the input sentence is input into the pre-trained reading understanding model, and a probability value of each part intercepted from the associated document as an answer sentence is predicted, specifically including:
in the pre-training stage, covering partial words of each associated document, inputting the partial words into a pre-trained reading understanding model to predict the covered partial words, and obtaining word vectors, position information of the word vectors and semantic information of the word vectors of each participle in the question sentences and the associated documents;
in the reading and understanding stage, word vectors of each participle in question sentences and associated documents, position information of the word vectors and semantic information of the word vectors are coded and input into a pre-trained reading and understanding model to predict probability values of each part intercepted from the associated documents as answer sentences.
Further, in the reading understanding phase, the encoding of the word vector of each participle in the question sentence and the associated document, the position information of the word vector, and the semantic information of the word vector is input to a pre-trained reading understanding model to predict the probability value of each part intercepted from the associated document as an answer sentence, and specifically includes:
in the reading and understanding stage, coding word vectors, position information of the word vectors and semantic information of the word vectors of each participle in the question sentences and the associated documents to obtain word codes and position codes;
inputting the operation result between the word code and the position code into a pre-trained reading understanding model to enable position information to be supplemented into the word code, and acquiring the association relation between the question sentence and each part intercepted from the associated document;
and predicting probability values of the parts intercepted from the associated documents as answer sentences based on the association relationship between the question sentences and the parts intercepted from the associated documents.
Further, the generating an output answer sentence by using each part intercepted from the associated document as a probability value of the answer sentence specifically includes:
sorting probability values of all parts intercepted from the associated documents as answer sentences according to the screening instruction;
and acquiring each part intercepted from the associated document as a partial document with the highest answer sentence probability value to generate an answer sentence.
According to another aspect of the present invention, there is provided an intelligent question answering device, comprising:
the acquisition unit is used for acquiring a related document, the matching degree of which with the question statement is ranked before a preset numerical value, from a pre-sorted knowledge base when the question statement is received;
the prediction unit is used for forming an input sentence by each part intercepted from each associated document and the question sentence, inputting the input sentence into a pre-trained reading understanding model, and predicting the probability value of each part intercepted from each associated document as an answer sentence;
and the generating unit is used for generating the output answer sentence by using each part intercepted from the associated document as the probability value of the answer sentence.
Further, the document set sorted from each website is recorded in the pre-sorted knowledge base, and the apparatus further includes:
the establishing unit is used for performing word segmentation processing on a document set in a pre-sorted knowledge base before acquiring associated documents with the problem statement matching degree ranked before a preset numerical value from the pre-sorted knowledge base when the problem statement is received, and establishing an inverted index from word segmentation to each document in the document set;
and the counting unit is used for carrying out word frequency counting on the participles of each document in the document set by using the word bag model to obtain the word frequency of the participles in each document.
Further, the acquisition unit includes:
the calculation module is used for calculating the evaluation value of the importance degree of the question statement in each document based on the inverted index of each document in the document set from the word segmentation;
and the selecting module is used for sorting according to the evaluation values from large to small and selecting the document with the evaluation value ranking before the preset value as the associated document.
Further, the calculation module is specifically configured to perform word segmentation processing on the question statement, and query word frequencies of the respective words in the question statement in the respective documents in the document set and word segmentation amounts contained in the respective documents based on the inverted indexes of the respective documents in the document set from the words;
the calculation module is specifically configured to calculate an evaluation value of the importance degree of each participle in the question sentence in each document according to the number of documents in the document set, the word frequency of each participle in the question sentence in each document in the document set, and the amount of the participles contained in each document;
the calculation module is specifically configured to summarize the evaluation values of the importance degrees of the respective participles in the question sentence in the respective documents, and obtain the evaluation values of the importance degrees of the question sentence in the respective documents.
Further, the pre-trained reading understanding model performs fine-tune training and prediction of reading understanding task on the question and answer data set by using a bert pre-training model, including a pre-training phase and a reading understanding phase, and the prediction unit includes:
the first prediction module is used for covering partial words of each associated document in a pre-training stage, inputting the partial words into a pre-trained reading understanding model to predict the covered partial words, and obtaining word vectors of each participle in the question sentences and the associated documents, position information of the word vectors and semantic information of the word vectors;
and the second prediction module is used for coding the word vector of each participle in the question sentence and the associated document, the position information of the word vector and the semantic information of the word vector in the reading understanding stage, and inputting the coded word vector, the position information of the word vector and the semantic information of the word vector into a pre-trained reading understanding model to predict the probability value of each part intercepted from the associated document as an answer sentence.
Further, the second prediction module is specifically configured to, in a reading understanding stage, encode a word vector, position information of the word vector, and semantic information of the word vector of each participle in the question statement and the associated document to obtain a word code and a position code;
the second prediction module is specifically configured to input an operation result between the word code and the position code to a pre-trained reading understanding model so that position information is supplemented to the word code, and obtain an association relationship between the question statement and each part intercepted from the associated document;
the second prediction module is specifically configured to predict probability values of the parts intercepted from the associated document as answer sentences based on the association relationship between the question sentences and the parts intercepted from the associated document.
Further, the generation unit includes:
the sorting module is used for sorting probability values of all parts intercepted from the associated documents as answer sentences according to the screening instructions;
and the generating module is used for acquiring each part intercepted from the associated document as a part document with the highest answer sentence probability value to generate the answer sentence.
According to yet another aspect of the present invention, there is provided a computer device comprising a memory storing a computer program and a processor implementing the steps of the intelligent question-answering method when the processor executes the computer program.
According to a further aspect of the present invention, there is provided a computer storage medium having a computer program stored thereon, which, when executed by a processor, performs the steps of the intelligent question-answering method.
By means of the technical scheme, when question sentences are received, relevant documents with the matching degree of the question sentences ranked before preset values are obtained from a pre-arranged knowledge base, and further, all parts intercepted from the relevant documents and the question sentences form an input sentence which is input into a pre-trained reading understanding model, and the probability values of all parts intercepted from the relevant documents are predicted to serve as answer sentences, so that output answer sentences are generated. Compared with the intelligent question-answering method in the prior art, the document sets arranged from various websites are recorded in the knowledge base arranged in advance, a more complete question-answering database is provided, the question sentences input by the user can be understood by utilizing the pre-trained reading understanding model, the retrieval intention of the user is analyzed, the probability value serving as answer sentences in the associated documents is predicted, the high-quality answer sentences of the user are given, and the accuracy of the generated question-answering is improved.
Drawings
Various other advantages and benefits will become apparent to those of ordinary skill in the art upon reading the following detailed description of the preferred embodiments. The drawings are only for purposes of illustrating the preferred embodiments and are not to be construed as limiting the invention. Also, like reference numerals are used to refer to like parts throughout the drawings. In the drawings:
fig. 1 is a schematic flow chart illustrating an intelligent question answering method according to an embodiment of the present invention;
FIG. 2 is a flow chart of another intelligent question answering method according to an embodiment of the present invention;
FIG. 3 is a diagram illustrating a process of emotion recognition on a question-answer corpus according to an embodiment of the present invention;
fig. 4 is a schematic structural diagram illustrating an intelligent question answering device according to an embodiment of the present invention;
fig. 5 is a schematic structural diagram of another intelligent question answering device according to an embodiment of the present invention.
Detailed Description
Exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be embodied in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.
The embodiment of the invention provides an intelligent question-answering method, which can understand question sentences input by a user, analyze the search intention of the user and give high-quality answer sentences to the user, and as shown in figure 1, the method comprises the following steps:
101. and when a question statement is received, acquiring a related document with the matching degree of the question statement ranked before a preset numerical value from a pre-sorted knowledge base.
The pre-arranged knowledge base can be a Wikipedia knowledge base, the Wikipedia knowledge base is a website similar to Baidu, the website content can be arranged into an open domain knowledge base, and the knowledge base comprises various documents and data sets and is downloaded and used during training of various artificial intelligence algorithms.
It can be understood that because the open questions in the existing question-answering field are too wide and the timeliness required by topics is high, the fixed corpus cannot meet the requirement of question-answering, the problem that the search speed and the understanding accuracy of question sentences are considered when the answer sentences are searched in massive question sentences is solved by adopting the Wikipedia knowledge base as the question-answering corpus, and a feasible idea is provided for the intelligent question-answering in the open field.
Specifically, when the associated documents ranked before the preset value with the matching degree of the question statement are acquired from the pre-sorted knowledge base, because the document sets sorted from the websites are recorded in the pre-sorted knowledge base, the documents ranked before the preset value with the importance degree can be selected as the associated documents according to the importance degree of the question statement in each document in the document set, and certainly, the documents ranked before the preset value with the frequency can be selected as the associated documents according to the frequency of the question statement in each document in the document set, which is not limited here.
102. And combining each part intercepted from each associated document and the question sentence into an input sentence, inputting the input sentence into a pre-trained reading understanding model, and predicting the probability value of each part intercepted from each associated document as an answer sentence.
Wherein the pre-trained reading understanding model performs a fine-tune training and prediction of the reading understanding task on the question-answer dataset by using a bert pre-training model. The method is characterized in that a bert pre-training model is used as a language model using a bidirectional transformer structure, and comprises the steps of pre-training in a pre-training stage and a reading understanding stage, training two types of tasks, wherein one type is to put together 15% of words in a document and predict the words of the masks in the training process; another category is to predict whether the next sentence in a sentence pair is the next sentence of the previous sentence.
And predicting each part intercepted from the relevant document according to the word vector of each word in the question sentence and the relevant document, the position information of the word vector and the semantic information of the word, so as to obtain the probability value of each part intercepted from the relevant document as an answer sentence.
103. And generating the output answer sentence by using each part intercepted from the associated document as the probability value of the answer sentence.
Because the associated text is a document which is retrieved after the text tracking and has a high degree of association with question sentences and is taken as question-answer corpus, the associated text is usually recorded with answer sentences, and the answer sentences in the associated document are predicted by reading an understanding model, so that the characteristics that a fixed corpus cannot meet the requirements due to high timeliness of topic requirements in open field questions are solved, the problems that the search speed and the accuracy of semantic understanding of the questions cannot be considered when the answers are searched in massive questions are solved, and the speed and the accuracy of question answering are improved.
According to the intelligent question-answering method provided by the embodiment of the invention, when a question sentence is received, an associated document with the matching degree of the question sentence ranked before a preset value is obtained from a pre-arranged knowledge base, each part intercepted from the associated document and the question sentence form an input sentence, the input sentence is input into a pre-trained reading understanding model, and the probability value of each part intercepted from the associated document is predicted to serve as an answer sentence, so that an output answer sentence is generated. Compared with the intelligent question-answering method in the prior art, the document sets arranged from various websites are recorded in the knowledge base arranged in advance, a more complete question-answering database is provided, the question sentences input by the user can be understood by utilizing the pre-trained reading understanding model, the retrieval intention of the user is analyzed, the probability value serving as answer sentences in the associated documents is predicted, the high-quality answer sentences of the user are given, and the accuracy of the generated question-answering is improved.
The embodiment of the invention provides another intelligent question-answering method, which can understand question sentences input by a user, analyze the search intention of the user and give high-quality answer sentences to the user, and as shown in fig. 2, the method comprises the following steps:
201. and performing word segmentation processing on the document set in the pre-sorted knowledge base, and establishing an inverted index from word segmentation to each document in the document set.
It can be understood that, for the documents in the pre-sorted knowledge base, in order to improve the document tracking efficiency, when the system is initialized, the word segmentation processing is firstly performed on each document in the document set in the pre-sorted knowledge base, and then the reverse index of the word segmentation to each document in the document set is established.
It should be noted that the present application is not limited to the word segmentation processing manner, and word segmentation tools such as a final word segmentation, LTP, HanLP, and the like may be used.
The documents and the participles in the document set are numbered by establishing the inverted indexes of the participles to the documents in the document set, so that the documents related to the problem sentences can be quickly found out from a mass document set based on the participle characteristics contained in the documents. For example, a document set includes 5 documents, and the document is subjected to word segmentation processing to obtain each word segmentation included in the document, each word segmentation has a corresponding number, the document number where the word segmentation appears is recorded, word a appears in documents 001 and 003, word B appears in document 004, word C appears in documents 001 and 004, word D appears in document 005, and the like, and accordingly, the inverted list corresponding to word a is {001 and 003}, the inverted list corresponding to word B is {004}, and the inverted list corresponding to word C is {001 and 004 }.
202. And performing word frequency statistics on the participles of each document in the document set by using a word bag model to obtain the word frequency of the participles in each document.
In which the Bag-of-words model was originally used in the field of information retrieval, for a document, it is assumed that the order relation and syntax of words in the document are not considered, and only whether the document has the word and the number of times the word appears (word frequency) are considered. Such a document is characterized by the words that appear in the document and the number of occurrences of each word.
For the embodiment of the present invention, specifically, after word frequency statistics may be performed on the segmented words of each document in the document set by using the word bag model to obtain the word frequency of each segmented word appearing in each document, the word frequency of each segmented word appearing in each document in the document set may be added to the inverted arrangement table of segmented words based on the inverted index of each document from the segmented word to the document set, for example, the inverted arrangement table corresponding to the segmented word a is {001, 003}, the number of times that the segmented word a appears in the document numbered 001 is 1 word, the number of times that the segmented word a appears in the document numbered 003 is 4 times, and accordingly, the inverted arrangement table of the segmented word a is updated to { (001; 1), (003; 4) }, so that the inverted arrangement table of the segmented word in each document is obtained, and each document where the segmented word appears and the word frequency in each document are recorded in the inverted arrangement table.
203. When a question sentence is received, based on the inverted indexes of the participles to the documents in the document set, the evaluation value of the importance degree of the question sentence in each document is calculated.
Specifically, in the process of calculating the evaluation value of the importance degree of the question sentence in each document based on the inverted index of each document from the participle to the document set, the participle included in the question sentence is obtained by performing participle processing on the question sentence, the word frequency of the participle included in the question sentence in each document is obtained based on the established inverted index of each document from the participle to the document set, and the evaluation value of the importance degree of the participle included in the question sentence in each document is further calculated.
For the embodiment of the present invention, the evaluation value for calculating the importance degree of the participles contained in the problem statement in each document can be obtained by calculating tf-idf values of the participles contained in the problem statement in each document, where tf-idf is a common weighting technique used for information retrieval and text mining. tf-idf is also a statistical method to evaluate how important a word is to one of the documents in a corpus or a corpus. The importance of a word is proportional to the number of times it appears in the document, but at the same time decreases inversely as the frequency with which it appears in the corpus.
Specifically, in the process of calculating the evaluation value of the importance degree of the problem statement in each document based on the inverted index from the participle to each document in the document set, the document number in the document set, the word frequency of each participle in the problem statement in each document in the document set and the participle amount contained in each document are recorded in the inverted index, and the tf-idf value of each participle in the problem statement in each document is calculated according to the document number in the document set, the word frequency of each participle in the problem statement in each document in the document set and the participle amount contained in each document; and then summarizing tf-idf values of all the participles in the question sentences in all the documents to obtain tf-idf values of the question sentences in all the documents.
Specifically, in the process of calculating tf-idf values of all participles in problem sentences in all documents, the word frequency termFreq of the participles in the problem sentences appearing in each document is firstly acquired, and then the total participle number docotota appearing in each document is acquiredlTerm; then
Acquiring the total document number as docNum and the document number containing the participles in the problem sentence as wordlndoccum, wherein idf is 1.0+ log (docNum/(wordlndoccum + 1)); tf-idf ═ tf × idf for each participle in the question sentence; the tf-idf of a question statement is the sum of the tf-idf of all the participles in the question statement/number of participles of the question statement.
Illustratively, the question sentence is divided into a participle a, a participle B and a participle C, wherein the participle a has a term frequency termFreq of 10 in the document 1, and the participle a has a term frequency termFreq of 7 in the document 2; the word frequency termFreq of the participle B appearing in the document 1 is 5, and the word frequency termFreq of the participle B appearing in the document 2 is 0; the word frequency termFreq of the participle C appearing in the document 1 is 20, and the word frequency termFreq of the participle C appearing in the document 2 is 10; the number of participles in document 1 is 100, and the number of participles in document 2 is 140; the participle A is in the document 1
Word segmentation A in document 2
With word segmentation B in document 1
With word segmentation B in document 2
Word segmentation C in document 1
Word segmentation C in document 2
Of the word segmentation A
Idf of the participle B is 1.0+ log (2/(2+ 1)); idf of the participle C is 1.0+ log (2/(1+ 1)); then
Finally, the tf-idf value of the question sentence in the document 1 is calculated, namely the tf-idf value of the participle A in the document 1, the tf-idf value of the participle B in the document 1 and the tf-idf value of the participle C in the document 1 are calculated; the tf-idf value of the computational problem statement in document 2-tf-idf value of the participle a in document 2+ tf-idf value of the participle B in document 2+ tf-idf value of the participle C in document 2/3.
204. And sorting the evaluation values from large to small, and selecting the document with the evaluation value ranking before a preset numerical value as an associated document.
Since the larger the tf-idf value is, the more important the document is for the question sentence, the higher the matching degree with the question sentence is, the larger the evaluation value is, the higher the association degree between the question sentence and the document is, and thus the document with the evaluation value ranking before the preset value is selected as the associated document.
It will be appreciated that if too many associated documents are selected and the subsequent recognition workload is also excessive, the answer speed of the intelligent dialog will be affected, and the preferred number of associated documents is 5 to 10.
205. In the pre-training stage, part of words of each associated document are covered, the covered part of words are input into a pre-trained reading understanding model to predict, and word vectors, position information of the word vectors and semantic information of the word vectors of each participle in the question sentences and the associated documents are obtained.
For the embodiment of the invention, in the pre-training stage, the bert pre-training model shields partial words in each associated document, and then predicts the original semantic information of the participle by using the context information of the participle, so that the learned semantic information can be fused with the context information of the left side and the right side of the participle, and further extracts the word vector, the position information of the word vector and the semantic information of the word vector of each participle in the problem statement and the associated document.
206. In the reading and understanding stage, word vectors of each participle in question sentences and associated documents, position information of the word vectors and semantic information of the word vectors are coded and input into a pre-trained reading and understanding model to predict probability values of each part intercepted from the associated documents as answer sentences.
For the embodiment of the invention, in the reading understanding stage, a word code and a position code are obtained by encoding a word vector, position information of the word vector and semantic information of the word vector of each participle in a question sentence and an associated document, an operation result between the word code and the position code is input into a pre-trained reading understanding model to enable the position information to be supplemented into the word code, an association relation between the question sentence and each part intercepted from the associated document is obtained, and a probability value of each part intercepted from the associated document as an answer sentence is predicted based on the association relation between the question sentence and each part intercepted from the associated document.
Specifically, in question sentences and associated documents, a process of encoding word vectors of each participle, position information of the word vectors and semantic information of the word vectors in the question sentences and the associated documents is carried out, wherein the word vectors are vectors of which each participle corresponds to one 768-dimension; the position information is an integer number which is marked for each participle in advance, and is converted into a vector with 768 dimensions according to the integer number; the semantic information is that question sentences are distinguished from associated documents in a reading understanding model, participles in all question sentences are marked as 0, participles in all associated documents are marked as 1, then 0 and 1 are converted into vectors with 768 dimensions, the input of the reading understanding model trained in advance in a reading understanding phase is the addition of word vectors, position vectors and semantic vectors, answer sentences corresponding to the question sentences are a section of text intercepted from the associated documents, the start positions start-point of the answer sentences in the associated documents and the end positions end-point of the answer sentences in the associated documents are assumed, and the probability values of taking each participle in each part intercepted from the documents as start-point and end-point can be predicted through the reading understanding model trained in advance.
207. And sorting probability values of all parts intercepted from the associated documents as answer sentences according to the screening instruction.
It can be understood that, in general, the higher the probability value of each part intercepted from the associated document as an answer sentence is, the more suitable the content of the part is as the answer sentence, and the part intercepted as the answer sentence with the highest probability value can be selected to generate the output answer sentence, thereby providing more accurate answer content for the user.
Further, in order to improve the flexibility of outputting the answer sentences, when the user inputs question sentences, the scene factors such as topic practical application occasions and contexts which may be considered by the user, and the documents with the highest probability values of the answer sentences which are obtained by intercepting all parts from the associated documents do not meet the scene factors, so before the output answer sentences are generated, the probability values of the answer sentences which are obtained by intercepting all parts from the associated documents can be sorted by setting a screening instruction and combining the current scene factors of the user, and therefore, more suitable parts of the documents are selected to generate the output answer sentences, and the scene factors are not limited.
208. And acquiring each part intercepted from the associated document as a partial document with the highest answer sentence probability value to generate an answer sentence.
For the embodiment of the present invention, as shown in fig. 3, when inputting a question sentence of a user, a document associated with the question sentence is tracked in real time from the wikipedia knowledge base, an associated document with the top rank of 5 is selected, and sentences in the question sentence and the associated document are input into a pre-trained reading understanding model for short text reading understanding, so that a sentence in the predicted document is used as a probability value of an answer sentence, and the answer sentence with the top probability value is selected as an optimal answer.
Further, as a specific implementation of the method shown in fig. 1, an embodiment of the present invention provides an intelligent question answering device, as shown in fig. 4, where the device includes: an acquisition unit 31, a prediction unit 32, and a generation unit 33.
The acquiring unit 31 may be configured to, when a question statement is received, acquire, from a knowledge base that is sorted in advance, an associated document whose matching degree with the question statement is ranked before a preset numerical value;
the prediction unit 32 may be configured to combine each part intercepted from each associated document with a question sentence to form an input sentence, input the input sentence into a pre-trained reading understanding model, and predict a probability value of each part intercepted from each associated document as an answer sentence;
the generating unit 33 may be configured to generate the output answer sentence by using the parts intercepted from the associated document as probability values of the answer sentence.
In the intelligent question answering device provided by the embodiment of the invention, when a question sentence is received, an associated document with a matching degree of the question sentence ranked before a preset value is obtained from a pre-arranged knowledge base, each part intercepted from the associated document and the question sentence form an input sentence, the input sentence is input into a pre-trained reading understanding model, and a probability value of each part intercepted from the associated document is predicted to serve as an answer sentence, so that an output answer sentence is generated. Compared with the intelligent question-answering method in the prior art, the document sets arranged from various websites are recorded in the knowledge base arranged in advance, a more complete question-answering database is provided, the question sentences input by the user can be understood by utilizing the pre-trained reading understanding model, the retrieval intention of the user is analyzed, the probability value serving as answer sentences in the associated documents is predicted, the high-quality answer sentences of the user are given, and the accuracy of the generated question-answering is improved.
As a further explanation of the intelligent question-answering device shown in fig. 4, fig. 5 is a schematic structural diagram of another intelligent question-answering device according to an embodiment of the present invention, as shown in fig. 5, in which document sets sorted from respective websites are recorded in the pre-sorted knowledge base, and the device further includes:
the establishing unit 34 may be configured to, when the question statement is received, perform word segmentation on a document set in a pre-sorted knowledge base before acquiring, from the pre-sorted knowledge base, an associated document whose matching degree with the question statement is ranked before a preset numerical value, and establish an inverted index from word segmentation to each document in the document set;
the statistical unit 35 may be configured to perform word frequency statistics on the segmented words of each document in the document set by using a word bag model, so as to obtain the word frequency of the segmented words appearing in each document.
In a specific application scenario, as shown in fig. 5, the obtaining unit 31 includes:
a calculating module 311, configured to calculate an evaluation value of the importance of the question and sentence in each document based on the inverted index of the participle to each document in the document set;
the selecting module 312 may be configured to sort the evaluation values from large to small, and select a document with the evaluation value ranking before a preset value as an associated document.
In a specific application scenario, the calculation module 311 may be specifically configured to perform word segmentation on the question statement, and query, based on the inverted index from the word segmentation to each document in the document set, a word frequency of each word in the question statement appearing in each document in the document set and a word segmentation amount included in each document;
the calculating module 311 may be further configured to calculate an evaluation value of an importance degree of each participle in the question sentence in each document according to the number of documents in the document set, a word frequency of each participle in the question sentence in each document in the document set, and a word quantity included in each document;
the calculating module 311 may be further configured to summarize the evaluation values of the importance degrees of the respective participles in the question sentence in the respective documents, so as to obtain the evaluation values of the importance degrees of the question sentence in the respective documents.
In a specific application scenario, as shown in fig. 5, the pre-trained reading understanding model performs fine-tune training and prediction of a reading understanding task on a question-answer data set by using a bert pre-training model, including a pre-training phase and a reading understanding phase, and the prediction unit 32 includes:
the first prediction module 321 may be configured to cover a part of words of each associated document in a pre-training stage, input the part of words into a pre-trained reading understanding model to predict the covered part of words, and obtain word vectors, position information of the word vectors, and semantic information of the word vectors of each participle in the question sentences and the associated documents;
the second prediction module 322 may be configured to, in the reading understanding phase, encode a word vector, position information of the word vector, and semantic information of the word vector of each participle in the question statement and the associated document, and input the encoded word vector, position information of the word vector, and a probability value that each part intercepted from the associated document is predicted as an answer statement to the pre-trained reading understanding model.
Further, the second prediction module 322 may be specifically configured to, in a reading understanding phase, encode a word vector, position information of the word vector, and semantic information of the word vector of each participle in the question statement and the associated document to obtain a word code and a position code;
the second prediction module 322 may be further configured to input the operation between the word code and the position code to a pre-trained reading understanding model so as to supplement the position information to the word code, and obtain an association relationship between the question statement and each part intercepted from the associated document;
the second prediction module 322 may be further configured to predict probability values of the parts intercepted from the associated document as answer sentences based on the association relationship between the question sentences and the parts intercepted from the associated document.
In a specific application scenario, as shown in fig. 5, the generating unit 33 includes:
the sorting module 331 may be configured to sort, according to the filtering instruction, probability values of the answer sentences of the respective parts intercepted from the associated document;
the generating module 332 may be configured to obtain each part intercepted from the associated document as a part document with the highest probability value of the answer sentence, and generate the answer sentence.
It should be noted that other corresponding descriptions of the functional units related to the intelligent question answering device provided in this embodiment may refer to the corresponding descriptions in fig. 1 to fig. 2, and are not described herein again.
Based on the above-mentioned methods shown in fig. 1 and fig. 2, correspondingly, the present embodiment further provides a storage medium, on which a computer program is stored, and the computer program, when being executed by a processor, implements the intelligent question answering method shown in fig. 1 and fig. 2.
Based on such understanding, the technical solution of the present application may be embodied in the form of a software product, which may be stored in a non-volatile storage medium (which may be a CD-ROM, a usb disk, a removable hard disk, etc.), and includes several instructions for enabling a computer device (which may be a personal computer, a server, or a network device, etc.) to execute the method according to the implementation scenarios of the present application.
Based on the method shown in fig. 1 and fig. 2 and the virtual device embodiment shown in fig. 4 and fig. 5, in order to achieve the above object, an embodiment of the present application further provides a computer device, which may specifically be a personal computer, a server, a network device, and the like, where the entity device includes a storage medium and a processor; a storage medium for storing a computer program; and a processor for executing a computer program to implement the intelligent question answering method shown in fig. 1 and 2.
Optionally, the computer device may also include a user interface, a network interface, a camera, Radio Frequency (RF) circuitry, sensors, audio circuitry, a WI-FI module, and so forth. The user interface may include a Display screen (Display), an input unit such as a keypad (Keyboard), etc., and the optional user interface may also include a USB interface, a card reader interface, etc. The network interface may optionally include a standard wired interface, a wireless interface (e.g., a bluetooth interface, WI-FI interface), etc.
Those skilled in the art will appreciate that the physical device structure of the intelligent question answering device provided in the present embodiment does not constitute a limitation to the physical device, and may include more or less components, or combine some components, or arrange different components.
The storage medium may further include an operating system and a network communication module. The operating system is a program that manages the hardware and software resources of the computer device described above, supporting the operation of information handling programs and other software and/or programs. The network communication module is used for realizing communication among components in the storage medium and other hardware and software in the entity device.
Through the above description of the embodiments, those skilled in the art will clearly understand that the present application can be implemented by software plus a necessary general hardware platform, and can also be implemented by hardware. By applying the technical scheme of the application, compared with the prior art, the document sets arranged from all websites are recorded in the knowledge base arranged in advance, a more complete question-answer database is provided, the question sentences input by the user can be understood by using the pre-trained reading understanding model, the retrieval intention of the user is analyzed, the probability value of the answer sentences in the associated documents is predicted, the high-quality answer sentences of the user are provided, and the accuracy of the generating question-answer is improved.
Those skilled in the art will appreciate that the figures are merely schematic representations of one preferred implementation scenario and that the blocks or flow diagrams in the figures are not necessarily required to practice the present application. Those skilled in the art will appreciate that the modules in the devices in the implementation scenario may be distributed in the devices in the implementation scenario according to the description of the implementation scenario, or may be located in one or more devices different from the present implementation scenario with corresponding changes. The modules of the implementation scenario may be combined into one module, or may be further split into a plurality of sub-modules.
The above application serial numbers are for description purposes only and do not represent the superiority or inferiority of the implementation scenarios. The above disclosure is only a few specific implementation scenarios of the present application, but the present application is not limited thereto, and any variations that can be made by those skilled in the art are intended to fall within the scope of the present application.
Claims (10)
1. An intelligent question-answering method, characterized in that the method comprises:
when a question statement is received, acquiring a related document with the matching degree of the question statement ranked before a preset numerical value from a pre-arranged knowledge base;
combining each part intercepted from each associated document and question sentences into an input sentence, inputting the input sentence into a pre-trained reading understanding model, and predicting the probability value of each part intercepted from each associated document as an answer sentence;
and generating the output answer sentence by using each part intercepted from the associated document as the probability value of the answer sentence.
2. The method according to claim 1, wherein the pre-arranged knowledge base records document sets arranged from various websites, and before acquiring, when a question statement is received, an associated document with a matching degree of the question statement ranking before a preset value from the pre-arranged knowledge base, the method further comprises:
performing word segmentation processing on the document set in the pre-sorted knowledge base, and establishing an inverted index from word segmentation to each document in the document set;
and performing word frequency statistics on the participles of each document in the document set by using a word bag model to obtain the word frequency of the participles in each document.
3. The method according to claim 2, wherein the obtaining of the associated document ranked before the preset numerical value with the matching degree of the question statement from the pre-sorted knowledge base specifically comprises:
calculating evaluation values of importance degrees of the problem sentences in all the documents based on the inverted indexes of the documents in the document set from the word segmentation;
and sorting the evaluation values from large to small, and selecting the document with the evaluation value ranking before a preset numerical value as an associated document.
4. The method according to claim 3, wherein the calculating an evaluation value of the degree of importance of the question sentence in each document based on the inverted index of the participle to each document in the document set specifically comprises:
performing word segmentation processing on the question sentences, and inquiring word frequency of each word in the question sentences in each document in the document set and word segmentation amount contained in each document based on the inverted indexes of each document in the document set from the word segmentation;
calculating the evaluation value of the importance degree of each participle in the question sentence in each document according to the number of documents in the document set, the word frequency of each participle in the question sentence in each document in the document set and the word quantity contained in each document;
and summarizing the evaluation value of the importance degree of each participle in the question sentence in each document to obtain the evaluation value of the importance degree of the question sentence in each document.
5. The method according to claim 1, wherein the pre-trained reading understanding model performs fine-tune training and prediction of reading understanding task on the question-answer data set by using a bert pre-training model, and comprises a pre-training phase and a reading understanding phase, wherein the parts intercepted from each associated document and question sentences form an input sentence, the input sentence is input into the pre-trained reading understanding model, and the parts intercepted from the associated document are predicted as probability values of answer sentences, and the method specifically comprises the following steps:
in the pre-training stage, covering partial words of each associated document, inputting the partial words into a pre-trained reading understanding model to predict the covered partial words, and obtaining word vectors, position information of the word vectors and semantic information of the word vectors of each participle in the question sentences and the associated documents;
in the reading and understanding stage, word vectors of each participle in question sentences and associated documents, position information of the word vectors and semantic information of the word vectors are coded and input into a pre-trained reading and understanding model to predict probability values of each part intercepted from the associated documents as answer sentences.
6. The method according to claim 5, wherein in the reading understanding phase, the word vector, the position information of the word vector, and the semantic information of the word vector of each participle in the question sentence and the associated document are encoded and input to a pre-trained reading understanding model to predict a probability value of each part intercepted from the associated document as an answer sentence, and specifically includes:
in the reading and understanding stage, coding word vectors, position information of the word vectors and semantic information of the word vectors of each participle in the question sentences and the associated documents to obtain word codes and position codes;
inputting the operation result between the word code and the position code into a pre-trained reading understanding model to enable position information to be supplemented into the word code, and acquiring the association relation between the question sentence and each part intercepted from the associated document;
and predicting probability values of the parts intercepted from the associated documents as answer sentences based on the association relationship between the question sentences and the parts intercepted from the associated documents.
7. The method according to any one of claims 1 to 6, wherein the generating an output answer sentence by using the probability values of the parts intercepted from the associated document as answer sentences specifically comprises:
sorting probability values of all parts intercepted from the associated documents as answer sentences according to the screening instruction;
and acquiring each part intercepted from the associated document as a partial document with the highest answer sentence probability value to generate an answer sentence.
8. An intelligent question answering device, characterized in that the device comprises:
the acquisition unit is used for acquiring a related document, the matching degree of which with the question statement is ranked before a preset numerical value, from a pre-sorted knowledge base when the question statement is received;
the prediction unit is used for forming an input sentence by each part intercepted from each associated document and the question sentence, inputting the input sentence into a pre-trained reading understanding model, and predicting the probability value of each part intercepted from each associated document as an answer sentence;
and the generating unit is used for generating the output answer sentence by using each part intercepted from the associated document as the probability value of the answer sentence.
9. A computer device comprising a memory and a processor, the memory storing a computer program, wherein the processor implements the steps of the method of any one of claims 1 to 7 when executing the computer program.
10. A computer storage medium on which a computer program is stored, characterized in that the computer program, when being executed by a processor, carries out the steps of the method of any one of claims 1 to 7.
Priority Applications (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010091180.4A CN111368042A (en) | 2020-02-13 | 2020-02-13 | Intelligent question and answer method and device, computer equipment and computer storage medium |
| PCT/CN2020/092963 WO2021159632A1 (en) | 2020-02-13 | 2020-05-28 | Intelligent questioning and answering method and apparatus, computer device, and computer storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010091180.4A CN111368042A (en) | 2020-02-13 | 2020-02-13 | Intelligent question and answer method and device, computer equipment and computer storage medium |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN111368042A true CN111368042A (en) | 2020-07-03 |
Family
ID=71206240
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010091180.4A Pending CN111368042A (en) | 2020-02-13 | 2020-02-13 | Intelligent question and answer method and device, computer equipment and computer storage medium |
Country Status (2)
| Country | Link |
|---|---|
| CN (1) | CN111368042A (en) |
| WO (1) | WO2021159632A1 (en) |
Cited By (15)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112100326A (en) * | 2020-08-28 | 2020-12-18 | 广州探迹科技有限公司 | Anti-interference knowledge base question-answering method and system integrating retrieval and machine reading understanding |
| CN112287085A (en) * | 2020-11-06 | 2021-01-29 | 中国平安财产保险股份有限公司 | Semantic matching method, system, device and storage medium |
| CN112347223A (en) * | 2020-11-03 | 2021-02-09 | 平安科技（深圳）有限公司 | Document retrieval method, document retrieval equipment and computer-readable storage medium |
| 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 |
| CN112685538A (en) * | 2020-12-30 | 2021-04-20 | 北京理工大学 | Text vector retrieval method combined with external knowledge |
| CN112800202A (en) * | 2021-02-05 | 2021-05-14 | 北京金山数字娱乐科技有限公司 | Document processing method and device |
| CN112883182A (en) * | 2021-03-05 | 2021-06-01 | 海信电子科技(武汉)有限公司 | Question-answer matching method and device based on machine reading |
| CN113704408A (en) * | 2021-08-31 | 2021-11-26 | 工银科技有限公司 | Retrieval method, retrieval apparatus, electronic device, storage medium, and program product |
| CN113934825A (en) * | 2021-12-21 | 2022-01-14 | 北京云迹科技有限公司 | Question answering method and device and electronic equipment |
| CN114444488A (en) * | 2022-01-26 | 2022-05-06 | 中国科学技术大学 | Reading understanding method, system, device and storage medium for few-sample machine |
| CN114780672A (en) * | 2022-03-23 | 2022-07-22 | 清华大学 | Medical question and answer processing method and device based on network resources |
| WO2022227165A1 (en) * | 2021-04-28 | 2022-11-03 | 平安科技（深圳）有限公司 | Question and answer method and apparatus for machine reading comprehension, computer device, and storage medium |
| WO2022252636A1 (en) * | 2021-06-01 | 2022-12-08 | 平安科技（深圳）有限公司 | Artificial intelligence-based answer generation method and apparatus, device, and storage medium |
| CN117474043A (en) * | 2023-12-27 | 2024-01-30 | 湖南三湘银行股份有限公司 | Intelligent question-answering system based on training model |
| WO2024066920A1 (en) * | 2022-09-30 | 2024-04-04 | 腾讯科技（深圳）有限公司 | Processing method and apparatus for dialogue in virtual scene, and electronic device, computer program product and computer storage medium |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20190108273A1 (en) * | 2017-10-10 | 2019-04-11 | Alibaba Group Holding Limited | Data Processing Method, Apparatus and Electronic Device |
| CN109918487A (en) * | 2019-01-28 | 2019-06-21 | 平安科技（深圳）有限公司 | Intelligent answer method and system based on network encyclopedia |
| CN110390003A (en) * | 2019-06-19 | 2019-10-29 | 北京百度网讯科技有限公司 | Question and answer processing method and system, computer equipment and readable medium based on medical treatment |
| CN110688491A (en) * | 2019-09-25 | 2020-01-14 | 暨南大学 | Machine reading understanding method, system, device and medium based on deep learning |
Family Cites Families (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106844368B (en) * | 2015-12-03 | 2020-06-16 | 华为技术有限公司 | Method for man-machine conversation, neural network system and user equipment |
| CN108446320A (en) * | 2018-02-09 | 2018-08-24 | 北京搜狗科技发展有限公司 | A kind of data processing method, device and the device for data processing |
| CN110309283B (en) * | 2019-06-28 | 2023-03-21 | 创新先进技术有限公司 | Answer determination method and device for intelligent question answering |
| CN110502621B (en) * | 2019-07-03 | 2023-06-13 | 平安科技（深圳）有限公司 | Question answering method, question answering device, computer equipment and storage medium |
-
2020
- 2020-02-13 CN CN202010091180.4A patent/CN111368042A/en active Pending
- 2020-05-28 WO PCT/CN2020/092963 patent/WO2021159632A1/en active Application Filing
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20190108273A1 (en) * | 2017-10-10 | 2019-04-11 | Alibaba Group Holding Limited | Data Processing Method, Apparatus and Electronic Device |
| CN109918487A (en) * | 2019-01-28 | 2019-06-21 | 平安科技（深圳）有限公司 | Intelligent answer method and system based on network encyclopedia |
| CN110390003A (en) * | 2019-06-19 | 2019-10-29 | 北京百度网讯科技有限公司 | Question and answer processing method and system, computer equipment and readable medium based on medical treatment |
| CN110688491A (en) * | 2019-09-25 | 2020-01-14 | 暨南大学 | Machine reading understanding method, system, device and medium based on deep learning |
Non-Patent Citations (2)
| Title |
|---|
| DANQI CHEN ET AL: "ReadingWikipedia to Answer Open-Domain Questions", 《ARXIV:1704.00051V2 [CS.CL]》, pages 1 - 10 * |
| WEI YANG ET AL: "End-to-End Open-Domain Question Answering with BERTserini", 《ARXIV:1902.01718V2 [CS.CL]》, pages 1 - 6 * |
Cited By (19)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112100326A (en) * | 2020-08-28 | 2020-12-18 | 广州探迹科技有限公司 | Anti-interference knowledge base question-answering method and system integrating retrieval and machine reading understanding |
| CN112347223A (en) * | 2020-11-03 | 2021-02-09 | 平安科技（深圳）有限公司 | Document retrieval method, document retrieval equipment and computer-readable storage medium |
| CN112347223B (en) * | 2020-11-03 | 2023-09-22 | 平安科技（深圳）有限公司 | Document retrieval method, apparatus, and computer-readable storage medium |
| CN112287085A (en) * | 2020-11-06 | 2021-01-29 | 中国平安财产保险股份有限公司 | Semantic matching method, system, device and storage medium |
| CN112287085B (en) * | 2020-11-06 | 2023-12-05 | 中国平安财产保险股份有限公司 | Semantic matching method, system, equipment and storage medium |
| 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 |
| CN112685538B (en) * | 2020-12-30 | 2022-10-14 | 北京理工大学 | Text vector retrieval method combined with external knowledge |
| CN112685538A (en) * | 2020-12-30 | 2021-04-20 | 北京理工大学 | Text vector retrieval method combined with external knowledge |
| CN112800202A (en) * | 2021-02-05 | 2021-05-14 | 北京金山数字娱乐科技有限公司 | Document processing method and device |
| CN112883182A (en) * | 2021-03-05 | 2021-06-01 | 海信电子科技(武汉)有限公司 | Question-answer matching method and device based on machine reading |
| WO2022227165A1 (en) * | 2021-04-28 | 2022-11-03 | 平安科技（深圳）有限公司 | Question and answer method and apparatus for machine reading comprehension, computer device, and storage medium |
| WO2022252636A1 (en) * | 2021-06-01 | 2022-12-08 | 平安科技（深圳）有限公司 | Artificial intelligence-based answer generation method and apparatus, device, and storage medium |
| CN113704408A (en) * | 2021-08-31 | 2021-11-26 | 工银科技有限公司 | Retrieval method, retrieval apparatus, electronic device, storage medium, and program product |
| CN113934825A (en) * | 2021-12-21 | 2022-01-14 | 北京云迹科技有限公司 | Question answering method and device and electronic equipment |
| CN114444488A (en) * | 2022-01-26 | 2022-05-06 | 中国科学技术大学 | Reading understanding method, system, device and storage medium for few-sample machine |
| CN114780672A (en) * | 2022-03-23 | 2022-07-22 | 清华大学 | Medical question and answer processing method and device based on network resources |
| WO2024066920A1 (en) * | 2022-09-30 | 2024-04-04 | 腾讯科技（深圳）有限公司 | Processing method and apparatus for dialogue in virtual scene, and electronic device, computer program product and computer storage medium |
| CN117474043A (en) * | 2023-12-27 | 2024-01-30 | 湖南三湘银行股份有限公司 | Intelligent question-answering system based on training model |
| CN117474043B (en) * | 2023-12-27 | 2024-04-02 | 湖南三湘银行股份有限公司 | Intelligent question-answering system based on training model |
Also Published As
| Publication number | Publication date |
|---|---|
| WO2021159632A1 (en) | 2021-08-19 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111368042A (en) | Intelligent question and answer method and device, computer equipment and computer storage medium | |
| CN110765247B (en) | Input prompting method and device for question-answering robot | |
| Penha et al. | Curriculum learning strategies for IR: An empirical study on conversation response ranking | |
| CN109325040B (en) | FAQ question-answer library generalization method, device and equipment | |
| CN112948534A (en) | Interaction method and system for intelligent man-machine conversation and electronic equipment | |
| CN113505204B (en) | Recall model training method, search recall device and computer equipment | |
| CN112100326B (en) | Anti-interference question and answer method and system integrating retrieval and machine reading understanding | |
| CN116911312B (en) | Task type dialogue system and implementation method thereof | |
| CN111767394A (en) | Abstract extraction method and device based on artificial intelligence expert system | |
| KR20200052412A (en) | Artificial intelligence employment system and employing method of thereof | |
| CN109710732A (en) | Information query method, device, storage medium and electronic equipment | |
| CN112905768A (en) | Data interaction method, device and storage medium | |
| CN114398473A (en) | Enterprise portrait generation method, device, server and storage medium | |
| CN110968664A (en) | Document retrieval method, device, equipment and medium | |
| Wei et al. | Online education recommendation model based on user behavior data analysis | |
| CN117149988B (en) | Data management processing method and system based on education digitization | |
| CN117520503A (en) | Financial customer service dialogue generation method, device, equipment and medium based on LLM model | |
| CN113159187A (en) | Classification model training method and device, and target text determining method and device | |
| CN113033912A (en) | Problem solving person recommendation method and device | |
| CN116881462A (en) | Text data processing, text representation and text clustering method and equipment | |
| CN108595395B (en) | Nickname generation method, device and equipment | |
| CN116401344A (en) | Method and device for searching table according to question | |
| CN114861657A (en) | Conference key sentence extraction method and device | |
| CN110659419B (en) | Method and related device for determining target user | |
| CN113761104A (en) | Method and device for detecting entity relationship in knowledge graph and electronic equipment |
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 |