CN109408619B - Method for dynamically calculating similarity between question and answer in question-answering field - Google Patents
Method for dynamically calculating similarity between question and answer in question-answering field Download PDFInfo
- Publication number
- CN109408619B CN109408619B CN201811177400.4A CN201811177400A CN109408619B CN 109408619 B CN109408619 B CN 109408619B CN 201811177400 A CN201811177400 A CN 201811177400A CN 109408619 B CN109408619 B CN 109408619B
- Authority
- CN
- China
- Prior art keywords
- question
- answer
- vector
- vectors
- calculating
- 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.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 32
- 239000013598 vector Substances 0.000 claims abstract description 121
- 230000011218 segmentation Effects 0.000 claims description 15
- 238000004364 calculation method Methods 0.000 claims description 13
- 238000012545 processing Methods 0.000 claims description 12
- 238000012549 training Methods 0.000 claims description 8
- 230000008569 process Effects 0.000 claims description 6
- 239000011159 matrix material Substances 0.000 claims description 5
- 230000004913 activation Effects 0.000 claims description 4
- 238000011176 pooling Methods 0.000 claims description 4
- 241000764238 Isis Species 0.000 claims description 2
- 230000006870 function Effects 0.000 description 8
- 238000011160 research Methods 0.000 description 8
- 238000010586 diagram Methods 0.000 description 4
- 238000007796 conventional method Methods 0.000 description 3
- 238000000605 extraction Methods 0.000 description 3
- 238000004458 analytical method Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 238000013473 artificial intelligence Methods 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000002457 bidirectional effect Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000005484 gravity Effects 0.000 description 1
- 238000003058 natural language processing Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention provides a method for dynamically calculating question and answer similarity in the field of question and answer, wherein the question comprises an answer question part and a question answer part, and the method for calculating the question and answer similarity comprises the following steps: calculating similarity scores of all vectors of answers of the answer question part and weighted question vectors, wherein the all vectors of the answers comprise answer entity vectors, answer type vectors and answer content vectors; and calculating the final similarity score of the question vector and the answer vector.
Description
Technical Field
The invention relates to the technical field of question answering, in particular to a method for dynamically calculating similarity between question sentences and answers in the field of question answering.
Background
In recent years, intelligent questioning and answering based on knowledge graph gradually becomes the research focus in the fields of artificial intelligence and natural language processing, and the mainstream research method comprises intelligent questioning and answering based on semantic analysis and intelligent questioning and answering based on information extraction. In the natural language question-answer research based on information extraction and combined with deep learning, research enthusiasm is being brought up at the present stage, and how to combine the relation between a question and an answer to calculate the similarity between the question and the answer is combined, so that a more accurate answer is returned to a model, and the center of gravity of the research is located.
At present, the research methods for natural language question answering based on knowledge graph at home and abroad can be divided into two categories: the first type is natural language question-answer based on semantic analysis, which converts natural language into corresponding logic form by using semantic analyzer, and then uses the generated result as structured query language (such as SPARQL query language) to search for correct answer in knowledge base; the second category is the knowledge-graph-based question-answer constructed by using neural networks, which is an information extraction-based method, which is currently taking the research of the hot tide. The knowledge graph is a huge graph in nature, and can also be called a knowledge base with a directed graph structure, namely a knowledge base of a semantic network. Nodes in the knowledge graph represent entities, and edges represent relationships between entities. In contrast to previous methods, knowledge-graph-based question-answering converts both questions and candidate answers into semantic vectors in a low-dimensional space, and one method is to find the correct answer that matches the question by a cosine-similarity type calculation between the question vector and the candidate answer vector. Secondly, the relevant knowledge of learning is represented by a knowledge graph. Typically a triplet consists of < entity, relationship, entity >, so if the correct entity and relationship are found, the correct answer can be found. Therefore, the method generally determines the main entity and the relationship in the question, finds the corresponding candidate main entity and the candidate relationship in the knowledge base, then finds the cosine similarity score between the candidate main entity in the knowledge base and the main entity in the question and the cosine similarity score between the candidate relationship in the knowledge base and the relationship in the question, and finally finds the correct answer by obtaining the highest candidate main entity and relationship.
Most knowledge-graph based question-answering studies today focus only on the impact of the question sheet aspects on the answers. However, for a question and answer, the question and answer are often not the same in the aspect of concern, e.g., the question "how many tickets are at the seven star park in Guilin? The question aspect tends to know the price of the ticket, while the answer aspect to find the correct answer to the question in the knowledge base first finds the main entity in the question, "septematic park", so the answer aspect is more concerned with the entity mentioned in the question, "septematic park". Therefore, to obtain more accurate answers, the two aspects need to be dynamically combined by considering both the question and answer factors. Therefore, the correct answer of a question and the answer thereof are also in inseparable connection, but the current knowledge graph-based question-answer research at home and abroad does not completely consider the correlation between the question and the answer in the aspect of calculating the similarity between the question and the answer, so that the answer is not accurate enough.
Disclosure of Invention
In view of the above-mentioned shortcomings of the prior art, an object of the present invention is to provide a method for dynamically calculating similarity between a question and an answer in a question-answering field, so as to solve the problem that the correlation between the question and the answer cannot be reflected when the similarity between the question and the answer is calculated in the current question-answering field.
In order to achieve the above and other related objects, the present invention provides a method for dynamically calculating question and answer similarity in question and answer field, wherein the question comprises an answer question portion and a question answer portion, and the method for calculating question and answer similarity comprises:
calculating similarity scores of all vectors of answers of the answer question part and weighted question vectors, wherein the all vectors of the answers comprise answer entity vectors, answer type vectors and answer content vectors;
and calculating the final similarity score of the question vector and the answer vector.
Optionally, the calculating a similarity score between the answer vector of the answer question portion and the weighted question vector specifically includes:
performing word segmentation processing on the Chinese question;
training the Chinese question after word segmentation processing by word2vec, and taking the trained vector as the input of Bi-LSTM to obtain the vector representation of the question;
calculating the weight of the question vector;
giving weight to the question vector;
and calculating similarity scores of all aspects of the question vectors and the answers.
Optionally, the calculating a final similarity score between the question vector and the answer vector specifically includes:
calculating average vector representation of different question sentences;
calculating the weight of the answer vector, namely the attention degree of different question sentences to all aspects of the answer;
and calculating the final similarity score of the question vector and the answer vector.
Optionally, the word segmentation processing on the chinese question specifically includes:
and performing word segmentation processing on the Chinese question by adopting Jieba word segmentation.
Optionally, the calculating an average vector representation of different question sentences specifically includes:
the vector output by the Bi-LSTM is connected into an average pooling layer to obtain a final vector representation;
in the formula,
is the average vector representation of different question sentences, n represents the number of word vectors contained in the divided question sentences, h representsjRepresents the vector representation of the question after Bi-LSTM treatment.
As described above, the method for dynamically calculating question-answer similarity in question-answer field according to the present invention has the following beneficial effects:
the invention provides a method for calculating the similarity between question and answer, which divides a question-answer model into two parts, namely an answer question part and a question answer part, calculates the similarity score between the first part of question and answer, and gives a weight to the first part through the second part to obtain the final similarity score.
Drawings
To further illustrate the description of the present invention, the following detailed description of the embodiments of the present invention is provided with reference to the accompanying drawings. It is appreciated that these drawings are merely exemplary and are not to be considered limiting of the scope of the invention.
FIG. 1 is a schematic flow chart of a method for dynamically calculating question and answer similarity according to the present invention;
FIG. 2 is a diagram of various aspects of question and answer attention in an example sentence;
FIG. 3 is a language question-answer of a conventional method;
FIG. 4 is a knowledge-graph based question-answer in accordance with the present invention.
Detailed Description
The embodiments of the present invention are described below with reference to specific embodiments, and other advantages and effects of the present invention will be easily understood by those skilled in the art from the disclosure of the present specification. The invention is capable of other and different embodiments and of being practiced or of being carried out in various ways, and its several details are capable of modification in various respects, all without departing from the spirit and scope of the present invention. It is to be noted that the features in the following embodiments and examples may be combined with each other without conflict.
It should be noted that the drawings provided in the following embodiments are only for illustrating the basic idea of the present invention, and the components related to the present invention are only shown in the drawings rather than drawn according to the number, shape and size of the components in actual implementation, and the type, quantity and proportion of the components in actual implementation may be changed freely, and the layout of the components may be more complicated.
In the existing technology for calculating similarity between question and answer, most methods only consider the single attention of the question to the answer and neglect the different attention of the answer to all aspects of the question, which can cause the found 'correct answer' to be too different from the result expected by people. The invention fully considers the influence of two aspects of questions and answers on selecting correct answers in the question-answering process, and divides a question-answering model into two parts: and the answer question part and the question answer part are answered, wherein the weight value obtained from the second part is assigned to the first part to obtain similarity scores of all aspects of the question and different aspects of the answer, so that the similarity score of the final question and the answer is obtained. FIG. 1 is a diagram illustrating the method for dynamically calculating the similarity between question and answer.
Firstly, considering an answer question part in a question-answer model, firstly, carrying out word segmentation processing on a Chinese question by adopting Jieba word segmentation for the processing of the question, then training the word2vec into a vector as the input of a bidirectional long-short term memory network (Bi-LSTM), obtaining the weight value of a question vector by a softmax function, and finally obtaining the similarity score of the question vector and the answer vector in all aspects (the answer is divided into three parts in the model, namely an answer entity, an answer type and answer content); when calculating different attention degrees of different questions of the second part to all aspects of the answer, accessing the vectors processed by the Bi-LSTM in the first part into an average pooling layer so as to obtain average vector representation of different question vectors, calculating the weight values of all aspects of the answer by using a softmax function in the first part, assigning the weight values to the similarity scores of the first part, and taking the candidate answer with the largest sum of the similarity scores as the final similarity score of the question vector and the answer vector and the final correct answer.
A method for dynamically calculating similarity between question sentences and answers in the question-answering field is disclosed, a flow diagram is shown in figure 1, and the method comprises the following steps:
step 1, calculating similarity scores of all vectors of answers of the partial answer question sentences and the weighted question sentence vectors. Wherein, dividing the answer into three parts: answer entity, answer type and answer content.
Step 1.1, for the processing of the question, firstly, word segmentation processing is carried out on the Chinese question by adopting Jieba word segmentation, and a vector after word2vec training after word segmentation is used as the input of Bi-LSTM processing, so that the vector representation of the obtained question is obtained:
wherein h isjRepresenting a Bi-LSTM-processed question vector representation comprising vectors in two directions
And
step 1.2, calculating the weight alpha of the question vectorijNamely, the attention degree of each aspect of the answer to different word vectors of the question sentence, the calculation formula is as follows:
wherein, wij＝f(WT[hj；ei]+b)，αijEssentially a softmax function, in which the parameter wijIs a non-linear activation function in which W isTIs the intermediate matrix of the answer question portion, b is the difference, and WTAnd b are updated randomly during the training process; in addition, ei∈(ee,et,ec) Aspects representing answers include, eeAnswer entity, etAnswer type and ecAnd (6) answer content.
Step 1.3, weighting is given to the question vector, and the weighted question vector is represented as qiThe calculation formula is as follows:
step 1.4, calculating similarity scores of all aspects of the question vectors and answers, wherein the calculation formula is as follows:
S(q,ei)＝f(qi,ei)
wherein the score function f (-) is the inner product of the question vector and the answer vector.
And 2, calculating the final similarity score of the question vector and the answer vector by combining the attention degrees of different question to all aspects of the answer in the question answer part.
Step 2.1 calculate the average vector representation of different question sentences
The vector output in the first part via Bi-LSTM is connected to an average pooling layer to obtain the final vector representation, which is calculated as:
in the formula, n represents the number of word vectors included in the question after word segmentation.
Step 2.2 calculate the weight beta of the answer vectoreiNamely the attention degree of different vectors to various aspects of the answer, the calculation formula is as follows:
wherein,
and is
Is the average vector representation of the different question sentences calculated in step 2.1. In the same way, in the formula
For softmax function, calculate
The function of (A) is a nonlinear activation function, WTAnd b are the intermediate matrix and the difference, respectively, whose values are updated randomly during the training process.
Step 2.3, calculating the final similarity score of the question vector and the answer vector, wherein the calculation formula is as follows:
in the formula, the weight of the answer vector is calculated
A similarity score S (q, e) given to each of the question vectors and answer vectors calculated in step 1i) In this step, the dynamic combination of question vectors and answer vectors is realized, so that the question and the answer are closely related.
And 3, combining the step 1 and the step 2, selecting the final correct answer of the question sentence, wherein the calculation formula is as follows:
Smax＝argmax{S(q,a)}(a∈Cq)
in the formula, CqRepresenting a set of candidate answers, argmax { f (x) } represents a variable x corresponding to making f (x) take the maximum value, so SmaxThe corresponding candidate answer vector is the correct answer of the final question.
When the similarity between the question and the answer is obtained by the method, the question and the answer are closely related by considering that the answer has different concerns about different word vectors of the question and different questions have different concerns about various aspects of the answer, and the correct answer obtained by using the model is more accurate and effective.
Fig. 2 is an exemplary diagram illustrating how a question sentence and an answer in an example sentence are different in terms of attention. From fig. 1, it can be seen that, for example, the example sentence "how many entrance tickets are at the seven star park in Guilin City? ". In the aspect of answer, to find the correct answer of the question in the knowledge base, the main entity 'Qixing park' in the question is firstly found, so that the entity 'Qixing park' mentioned in the question is more concerned in the aspect of answer; the question aspect is more likely to know the price of the entrance ticket, namely what the price is, and the attribute of the entity is concerned, so that the question aspect is different from the answer aspect.
Fig. 3 and 4 illustrate the difference between the linguistic questions and answers of the conventional method and the knowledge-graph-based questions and answers of the present invention. As can be seen from fig. 3, the question-answering uses a semantic parser to convert a natural language into a corresponding logical form, and then uses the generated result as a structured query language (e.g., SPARQL query language) to find a correct answer in a knowledge base. The question-answer based on the knowledge graph is, as shown in fig. 4, converted into semantic vectors of a low-dimensional space, and found out corresponding candidate answers in the knowledge base through the main entity in the question, and finally found out correct answers matched with the question through cosine similarity calculation between the question vectors and the candidate answer vectors.
The invention provides a method for dynamically calculating question and answer similarity in the field of question answering, which comprises the following steps: the model is divided into two parts, namely an answer question part and a question answer part, the answer vector weight calculated by the second part is assigned to the first part, namely the similarity score of the question vector and the answer vector in all aspects, the assigned similarity scores are summed up according to the three aspects of the answer, and the candidate answer corresponding to the maximum value of the sum of the final similarity scores is the correct answer of the question. The problem that the influence of only one side of a question on the selection of correct answers is concerned in the conventional method and more prepared answers cannot be obtained is solved. The method provided by the invention can more accurately calculate the similarity between the question and the answer and can be widely applied to the field of question answering.
The foregoing embodiments are merely illustrative of the principles and utilities of the present invention and are not intended to limit the invention. Any person skilled in the art can modify or change the above-mentioned embodiments without departing from the spirit and scope of the present invention. Accordingly, it is intended that all equivalent modifications or changes which can be made by those skilled in the art without departing from the spirit and technical spirit of the present invention be covered by the claims of the present invention.
Claims (1)
1. A method for dynamically calculating question and answer similarity facing to the question and answer field is characterized in that the question comprises an answer question part and a question answer part, and the method for calculating question and answer similarity comprises the following steps:
calculating similarity scores of all vectors of answers of the answer question part and weighted question vectors, wherein the all vectors of the answers comprise answer entity vectors, answer type vectors and answer content vectors;
calculating the final similarity score of the question vector and the answer vector, wherein the detailed process comprises the following steps:
step 1, calculating similarity scores of all vectors of answers of the question part of the answers and the weighted question vectors, wherein the answers are divided into three parts: answer entity, answer type and answer content;
step 1.1, for the processing of the question, firstly, word segmentation processing is carried out on the Chinese question by adopting Jieba word segmentation, and a vector after word2vec training after word segmentation is used as the input of Bi-LSTM processing, so that the vector representation of the obtained question is obtained:
wherein h isjRepresenting a Bi-LSTM-processed question vector representation comprising vectors in two directions
And
step 1.2, calculating the weight alpha of the question vectorijNamely, the attention degree of each aspect of the answer to different word vectors of the question sentence, the calculation formula is as follows:
wherein, wij＝f(WT[hj；ei]+b)，αijEssentially a softmax function, in which the parameter wijIs a non-linear activation function in which W isTIs the intermediate matrix of the answer question portion, b is the difference, and WTAnd b are updated randomly during the training process; in addition, ei∈(ee,et,ec) Aspects representing answers include, eeAnswer entity, etAnswer type and ecThe answer content;
step 1.3, weighting is given to the question vector, and the weighted question vector is represented as qiThe calculation formula is as follows:
step 1.4, calculating similarity scores of all aspects of the question vectors and answers, wherein the calculation formula is as follows:
S(q,ei)＝f(qi,ei)
wherein, the score function f (-) is the inner product of the question vector and the answer vector;
step 2, calculating final similarity scores of the question vectors and the answer vectors by combining attention degrees of different question portions of the question answers to all aspects of the answers;
step 2.1 calculate the average vector representation of different question sentences
The vector output in the first part via Bi-LSTM is connected to an average pooling layer to obtain the final vector representation, which is calculated as:
in the formula, n represents the number of word vectors contained in the question after word segmentation;
step 2.2 calculate the weight of the answer vector
Namely the attention degree of different vectors to various aspects of the answer, the calculation formula is as follows:
wherein,
and is
Is the average vector representation of the different question sentences calculated in step 2.1, in the same way, in the formula
For softmax function, calculate
The function of (A) is a nonlinear activation function, WTAnd b are the intermediate matrix and the difference respectively, and the values of the intermediate matrix and the difference are updated randomly in the training process;
step 2.3, calculating the final similarity score of the question vector and the answer vector, wherein the calculation formula is as follows:
in the formula, the weight of the answer vector is calculated
A similarity score S (q, e) given to each of the question vectors and answer vectors calculated in step 1i) The dynamic combination of the question vector and the answer vector makes the question and the answer closely related.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811177400.4A CN109408619B (en) | 2018-10-10 | 2018-10-10 | Method for dynamically calculating similarity between question and answer in question-answering field |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811177400.4A CN109408619B (en) | 2018-10-10 | 2018-10-10 | Method for dynamically calculating similarity between question and answer in question-answering field |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109408619A CN109408619A (en) | 2019-03-01 |
| CN109408619B true CN109408619B (en) | 2021-09-07 |
Family
ID=65467393
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811177400.4A Active CN109408619B (en) | 2018-10-10 | 2018-10-10 | Method for dynamically calculating similarity between question and answer in question-answering field |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109408619B (en) |
Families Citing this family (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111680133A (en) * | 2019-03-11 | 2020-09-18 | 阿里巴巴集团控股有限公司 | Live broadcast question and answer method and device |
| CN110727695B (en) * | 2019-09-29 | 2022-05-03 | 浙江大学 | Natural language query analysis method for novel power supply urban rail train data operation and maintenance |
| CN110866095A (en) * | 2019-10-10 | 2020-03-06 | 重庆金融资产交易所有限责任公司 | Text similarity determination method and related equipment |
| CN110825861B (en) * | 2019-11-05 | 2022-05-03 | 重庆邮电大学 | Man-machine conversation method and system based on distributed representation model confusion degree |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106227740A (en) * | 2016-07-12 | 2016-12-14 | 北京光年无限科技有限公司 | A kind of data processing method towards conversational system and device |
| CN106570708A (en) * | 2016-10-31 | 2017-04-19 | 厦门快商通科技股份有限公司 | Management method and management system of intelligent customer service knowledge base |
| EP3229157A1 (en) * | 2016-04-07 | 2017-10-11 | Siemens Healthcare GmbH | Image analytics question answering |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104572734B (en) * | 2013-10-23 | 2019-04-30 | 腾讯科技（深圳）有限公司 | Method for recommending problem, apparatus and system |
-
2018
- 2018-10-10 CN CN201811177400.4A patent/CN109408619B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP3229157A1 (en) * | 2016-04-07 | 2017-10-11 | Siemens Healthcare GmbH | Image analytics question answering |
| CN106227740A (en) * | 2016-07-12 | 2016-12-14 | 北京光年无限科技有限公司 | A kind of data processing method towards conversational system and device |
| CN106570708A (en) * | 2016-10-31 | 2017-04-19 | 厦门快商通科技股份有限公司 | Management method and management system of intelligent customer service knowledge base |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109408619A (en) | 2019-03-01 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109408619B (en) | Method for dynamically calculating similarity between question and answer in question-answering field | |
| CN108549658B (en) | Deep learning video question-answering method and system based on attention mechanism on syntax analysis tree | |
| CN102262634B (en) | Automatic questioning and answering method and system | |
| CN109271506A (en) | A kind of construction method of the field of power communication knowledge mapping question answering system based on deep learning | |
| CN111737426B (en) | Method for training question-answering model, computer equipment and readable storage medium | |
| CN110706807B (en) | Medical question-answering method based on ontology semantic similarity | |
| CN109783817A (en) | A kind of text semantic similarity calculation model based on deeply study | |
| CN106776711A (en) | A kind of Chinese medical knowledge mapping construction method based on deep learning | |
| WO2019165678A1 (en) | Keyword extraction method for mooc | |
| CN110309503A (en) | A kind of subjective item Rating Model and methods of marking based on deep learning BERT--CNN | |
| CN111897944B (en) | Knowledge graph question-answering system based on semantic space sharing | |
| CN112836027A (en) | Method for determining text similarity, question answering method and question answering system | |
| CN111581364B (en) | Chinese intelligent question-answer short text similarity calculation method oriented to medical field | |
| CN111143672A (en) | Expert specialty scholars recommendation method based on knowledge graph | |
| CN113342958B (en) | Question-answer matching method, text matching model training method and related equipment | |
| CN111666376B (en) | Answer generation method and device based on paragraph boundary scan prediction and word shift distance cluster matching | |
| CN112084312B (en) | Intelligent customer service system constructed based on knowledge graph | |
| CN115131698B (en) | Video attribute determining method, device, equipment and storage medium | |
| CN107491729A (en) | The Handwritten Digit Recognition method of convolutional neural networks based on cosine similarity activation | |
| CN112632250A (en) | Question and answer method and system under multi-document scene | |
| JP2023515902A (en) | Meta-knowledge fine-tuning method and platform for multitasking language models | |
| CN115345165A (en) | Specific entity identification method oriented to label scarcity or distribution unbalance scene | |
| CN113254609B (en) | Question-answering model integration method based on negative sample diversity | |
| CN110909175A (en) | Search engine-based online course concept knowledge graph construction method | |
| CN116842168B (en) | Cross-domain problem processing method and device, electronic equipment and storage medium |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| EE01 | Entry into force of recordation of patent licensing contract |
Application publication date: 20190301 Assignee: GUANGXI XINKAI ZHIQUAN TECHNOLOGY Co.,Ltd. Assignor: GUILIN University OF ELECTRONIC TECHNOLOGY Contract record no.: X2023980045064 Denomination of invention: A Method for Dynamically Calculating the Similarity between Questions and Answers in the Question and Answer Domain Granted publication date: 20210907 License type: Common License Record date: 20231101 |
|
| EE01 | Entry into force of recordation of patent licensing contract |