CN111782767A - Question and answer method, device, equipment and storage medium - Google Patents

Abstract

The present application discloses a question and answer method, apparatus, device and storage medium, which belong to the field of natural language processing. The method includes: calling an answer generation model, obtaining at least one piece of answer information according to the first question vector and at least one second question vector of the first question information, and searching in the question-and-answer database according to the first question information or the first question vector; The sorting model is invoked to sort the obtained multiple pieces of answer information in order of matching degree with the first question information from high to low, and the answer information ranked first is determined as the target answer information. It ensures that the answer information can be generated even if the answer information cannot be retrieved by the retrieval method, and there will be no missing answer information, and the influence of the context and the matching degree of the answer information and the question information are also considered comprehensively. Accuracy of answer information.

Description

Question and answer method, device, equipment and storage medium

technical field

The present application relates to the field of natural language processing, and in particular, to a question answering method, apparatus, device and storage medium.

Background technique

With the popularization of the Internet and the wide application of natural language processing technology, the intelligent question answering function has gradually emerged. After the user enters the question information, the intelligent question answering function can automatically answer the user's question, thereby interacting with the user.

In the related art, after a user inputs question information, a keyword in the question information is acquired, and based on the acquired keyword, answer information matching the keyword is retrieved in a database to complete the answer to the question information.

However, when the matching answer information cannot be retrieved in the database, the question information cannot be answered, so the above method has limitations.

SUMMARY OF THE INVENTION

The examples of the present application provide a question and answer method, apparatus, device and storage medium, which improve the accuracy of the obtained answer information. The technical solution is as follows:

In one aspect, a question answering method is provided, the method comprising:

acquiring a first question vector of first question information and at least one second question vector of second question information, where the at least one piece of second question information is question information obtained before the first question information;

Invoke the answer generation model, obtain at least one piece of answer information according to the first question vector and at least one second question vector, and the answer generation model is used to generate the answer information matched by any question vector according to any question vector;

According to the first question information or the first question vector, searching in the question answering database;

The sorting model is invoked to sort the obtained pieces of answer information in descending order of matching degree with the first question information, and the answer information ranked first is determined as the target answer information.

In a possible implementation, the method further includes:

Obtain sample question information and corresponding sample answer information, as well as the sample matching degree between the sample question information and the sample answer information;

According to the sample question information, the sample answer information and the sample matching degree, the ranking model is trained to obtain a trained ranking model.

In another possible implementation manner, the sorting model is trained according to the sample question information, the sample answer information and the sample matching degree to obtain a trained sorting model, including:

inputting the sample question information and the sample answer information into the ranking model, and obtaining the predicted matching degree between the sample question information and the sample answer information;

According to the sample matching degree and the predicted matching degree, the ranking model is trained to obtain a trained ranking model.

In another possible implementation manner, the ranking model includes a plurality of matching layers, a fusion layer and a ranking layer, and the ranking model is invoked in descending order of matching degree with the first question information , sort the multiple pieces of obtained answer information, and determine the first answer information as the target answer information, including:

Calling the multiple matching layers to obtain the matching degree of each piece of answer information and the first question information respectively;

Call the fusion layer to obtain the fusion matching degree of the multiple matching degrees of each piece of answer information;

The sorting layer is called to sort the multiple pieces of answer information according to the order of fusion matching degree from high to low, and the answer information ranked first is determined as the target answer information.

In another possible implementation manner, the obtaining the first question vector of the first question information includes:

Perform word segmentation processing on the first question information to obtain a plurality of first words after word segmentation;

obtaining similar words of the plurality of first words;

Each time at least one first word is replaced with a corresponding similar word to generate a piece of similar question information;

Obtain the first question vector according to the first question information and the at least one piece of similar question information.

In another possible implementation manner, the acquiring similar words of the plurality of first words includes:

Obtain, from a knowledge database, similar words associated with each of the plurality of first words, where the knowledge database is used to store similar words associated with each word; or,

Obtain the similarity between each first word in the plurality of first words and at least one preset word, and use a preset word whose similarity with any first word is greater than the first preset similarity as the arbitrary first word. A similar word to the first word.

In another possible implementation manner, the obtaining the first question vector according to the first question information and the at least one piece of similar question information includes:

An average vector of the feature vector of the at least one piece of similar question information and the feature vector of the first question information is acquired as the first question vector.

In another possible implementation manner, the invoking the answer generation model obtains at least one piece of answer information according to the first question vector and at least one second question vector, including:

splicing the first problem vector and the at least one second problem vector to obtain a third problem vector;

The answer generation model is called, and the at least one piece of answer information is acquired according to the third question vector.

In another possible implementation manner, the retrieval in the question and answer database according to the first question information or the first question vector includes:

splicing the first problem vector and the at least one second problem vector to obtain a third problem vector;

According to the similarity between the third question vector and the feature vector of each preset answer information in the question answering database, at least one piece of answer information is obtained, and the feature vector of the at least one piece of answer information is related to the third question The similarity of the vectors is greater than the similarity between the feature vectors of other preset answer information and the third question vector.

In another possible implementation manner, the at least one second question vector includes multiple second question vectors, and the answer generation model is invoked to obtain at least one second question vector according to the first question vector and the at least one second question vector Answer information, including:

performing a weighted average on the plurality of second problem vectors to obtain a fourth problem vector;

Splicing the first problem vector and the fourth problem vector to obtain a fusion problem vector;

According to the fusion question vector, the answer generation model is called to obtain the at least one piece of answer information.

In another possible implementation manner, the retrieval in the question and answer database according to the first question information or the first question vector includes:

performing a weighted average on the plurality of second problem vectors to obtain a fourth problem vector;

Splicing the first problem vector and the fourth problem vector to obtain a fusion problem vector;

At least one piece of answer information is obtained according to the similarity between the fusion question vector and the feature vector of each preset answer information in the question answering database, and the feature vector of the at least one piece of answer information is the difference between the fusion question vector and the fusion question vector. The similarity is greater than the similarity between the feature vector of other preset answer information and the fusion question vector.

In another possible implementation manner, the retrieval in the question and answer database according to the first question information or the first question vector includes:

According to the first question vector, perform retrieval in the question answering database to determine at least one fifth question vector, where the at least one fifth question vector includes question vectors similar to the first question vector; or,

According to the first question vector and the at least one second question vector, perform retrieval in the question answering database to determine at least one fifth question vector, the at least one fifth question vector including the first question vector is similar At least one of the problem vectors of or the problem vectors that are similar to the at least one second problem vector;

From the question and answer database, the answer information of the at least one fifth question vector is obtained as the at least one piece of answer information.

In another possible implementation manner, the retrieval in the question and answer database according to the first question information or the first question vector includes:

The at least one piece of answer information is acquired according to the similarity between the first question vector and the feature vector of each preset answer information in the question answering database, and the feature vector of the at least one piece of answer information is the same as the first piece of answer information. The similarity of a question vector is greater than the similarity between the feature vectors of other preset answer information and the first question vector.

In another possible implementation manner, the retrieval in the question and answer database according to the first question information or the first question vector includes:

Perform word segmentation processing on the first question information to obtain keywords or entities of the first question information;

Perform word segmentation processing on the preset answer information in the question and answer database to obtain keywords or entities of the preset answer information;

At least one piece of answer information is acquired according to the keywords or entities of the first question information and the keywords or entities of the preset answer information in the question-and-answer database.

In another aspect, a question and answer device is provided, the device comprising:

an acquisition module, configured to acquire the first question vector of the first question information and the second question vector of at least one piece of second question information, where the at least one piece of second question information is the question information acquired before the first question information;

The generation module is used to call the answer generation model to obtain at least one piece of answer information according to the first question vector and at least one second question vector, and the answer generation model is used to generate the any question vector according to any question vector matching answer information;

a retrieval module, configured to perform retrieval in the question and answer database according to the first question information or the first question vector;

The sorting module is used to call the sorting model, sort the obtained multiple pieces of answer information in the order of matching degree with the first question information from high to low, and determine the answer information ranked first as the Describe the target answer information.

In a possible implementation, the apparatus further includes:

the obtaining module, configured to obtain sample question information and corresponding sample answer information, and a sample matching degree between the sample question information and the sample answer information;

A training module, configured to train the ranking model according to the sample question information, the sample answer information and the sample matching degree to obtain a trained ranking model.

In another possible implementation, the training module includes:

an input unit, configured to input the sample question information and the sample answer information into the ranking model, and obtain the predicted matching degree between the sample question information and the sample answer information;

A training unit, configured to train the ranking model according to the sample matching degree and the predicted matching degree to obtain a trained ranking model.

In another possible implementation, the ranking model includes multiple matching layers, a fusion layer and a ranking layer, and the generation module includes:

a matching degree obtaining unit, configured to invoke the multiple matching layers to obtain the matching degree of each piece of answer information and the first question information respectively;

a fusion unit, used for calling the fusion layer to obtain a fusion matching degree of multiple matching degrees of each piece of answer information;

The sorting unit is used for invoking the sorting layer, sorting the multiple pieces of answer information according to the order of fusion matching degree from high to low, and determining the answer information ranked first as the target answer information.

In another possible implementation manner, the obtaining module is configured to execute any one of the following:

Perform word segmentation processing on the first question information to obtain a plurality of first words after word segmentation;

obtaining similar words of the plurality of first words;

Each time at least one first word is replaced with a corresponding similar word to generate a piece of similar question information;

Obtain the first question vector according to the first question information and the at least one piece of similar question information.

In another possible implementation manner, the obtaining module is configured to execute any one of the following:

Obtain, from a knowledge database, similar words associated with each of the plurality of first words, where the knowledge database is used to store similar words associated with each word; or,

Obtain the similarity between each first word in the plurality of first words and at least one preset word, and use a preset word whose similarity with any first word is greater than the first preset similarity as the arbitrary first word. A similar word to the first word.

In another possible implementation manner, the obtaining module is configured to obtain an average vector of the feature vector of the at least one piece of similar question information and the feature vector of the first question information, as the first question vector.

In another possible implementation manner, the generating module is configured to execute any one of the following:

splicing the first problem vector and the at least one second problem vector to obtain a third problem vector;

The answer generation model is called, and the at least one piece of answer information is acquired according to the third question vector.

In another possible implementation manner, the retrieval module is configured to perform any one of the following:

splicing the first problem vector and the at least one second problem vector to obtain a third problem vector;

According to the similarity between the third question vector and the feature vector of each preset answer information in the question answering database, at least one piece of answer information is obtained, and the feature vector of the at least one piece of answer information is related to the third question The similarity of the vectors is greater than the similarity between the feature vectors of other preset answer information and the third question vector.

In another possible implementation manner, the at least one second problem vector includes a plurality of second problem vectors, and the generating module is configured to perform any one of the following:

performing a weighted average on the plurality of second problem vectors to obtain a fourth problem vector;

Splicing the first problem vector and the fourth problem vector to obtain a fusion problem vector;

According to the fusion question vector, the answer generation model is called to obtain the at least one piece of answer information.

In another possible implementation manner, the retrieval module is configured to perform any one of the following:

performing a weighted average on the plurality of second problem vectors to obtain a fourth problem vector;

Splicing the first problem vector and the fourth problem vector to obtain a fusion problem vector;

At least one piece of answer information is obtained according to the similarity between the fusion question vector and the feature vector of each preset answer information in the question answering database, and the feature vector of the at least one piece of answer information is the difference between the fusion question vector and the fusion question vector. The similarity is greater than the similarity between the feature vector of other preset answer information and the fusion question vector.

In another possible implementation manner, the retrieval module is configured to perform any one of the following:

According to the first question vector, perform retrieval in the question answering database to determine at least one fifth question vector, where the at least one fifth question vector includes question vectors similar to the first question vector; or,

According to the first question vector and the at least one second question vector, searching in the question answering database to determine at least one fourth question vector, the at least one fourth question vector including the first question vector is similar At least one of the problem vectors of or the problem vectors that are similar to the at least one second problem vector;

From the question and answer database, the answer information of the at least one fourth question vector is obtained as the at least one piece of answer information.

In another possible implementation manner, the retrieval module is configured to perform any one of the following:

Perform word segmentation processing on the first question information to obtain keywords or entities of the first question information;

Perform word segmentation processing on the preset answer information in the question and answer database to obtain keywords or entities of the preset answer information;

At least one piece of answer information is acquired according to the keywords or entities of the first question information and the keywords or entities of the preset answer information in the question-and-answer database.

In another aspect, an electronic device is provided, the electronic device comprising one or more processors and one or more memories, the one or more memories having at least one instruction stored therein, the at least one instruction being The one or more processors are loaded and executed to implement operations as performed by the question answering method.

In another aspect, there is provided a computer-readable storage medium having stored therein at least one instruction, the at least one instruction being loaded and executed by a processor to implement operations performed by the question-and-answer method as described.

The question and answer method, device, device, and storage medium provided by the embodiments of the present application adopt the method of automatic generation and retrieval based on the model to obtain answer information, and then call the sorting model, according to the matching degree between the answer information and the question information. The answer information is sorted, and the answer information in the first place is obtained, which is the answer information that best matches the first question information, which ensures that the answer information can be generated even if the answer information cannot be retrieved by the retrieval method. There is no situation where the answer information is missing, and the influence of the context and the degree of matching between the answer information and the question information are comprehensively considered, and the accuracy of the answer information is improved.

Description of drawings

In order to illustrate the technical solutions in the embodiments of the present application more clearly, the following briefly introduces the drawings that are used in the description of the embodiments. Obviously, the drawings in the following description are only some embodiments of the present application. For those of ordinary skill in the art, other drawings can also be obtained from these drawings without creative effort.

1 is a schematic structural diagram of an implementation environment provided by an embodiment of the present application;

2 is a flowchart of a question-and-answer method provided by an embodiment of the present application;

3 is a flowchart of a question and answer method provided by an embodiment of the present application;

4 is a block diagram of an answer generation model provided by an embodiment of the present application;

5 is a block diagram of a retrieval answer information provided by an embodiment of the present application;

6 is a block diagram of sorting by a sorting model provided by an embodiment of the present application;

7 is a block diagram of a determination answer provided by an embodiment of the present application;

8 is a block diagram of operations performed by an input module provided by an embodiment of the present application;

9 is a schematic structural diagram of a question and answer device provided by an embodiment of the present application;

10 is a schematic structural diagram of a question and answer device provided by an embodiment of the present application;

FIG. 11 is a schematic structural diagram of a terminal provided by an embodiment of the present application;

FIG. 12 is a schematic structural diagram of a server provided by an embodiment of the present application.

Detailed ways

In order to make the objectives, technical solutions and advantages of the present application clearer, the embodiments of the present application will be further described in detail below with reference to the accompanying drawings.

The question-and-answer method provided by this application can automatically provide answers to the questions asked by users online, improve the accuracy of answering questions, and can be applied in various scenarios:

For example, the question-and-answer method provided in this application is applied in an intelligent customer service scenario. When a user uses any application, the application can provide the user with the service of the customer service Q&A function, and the user can enter question information in the application. , when the application detects the question information input by the user, and then adopts the question-and-answer method provided by the embodiment of the present application to determine a matching answer according to the question information, so that the user can view the determined answer.

Alternatively, the question-and-answer method provided by the present application is applied in an entertainment interaction scenario. When the user needs to have entertainment, he can interact with the intelligent robot, and the user can input question information to the intelligent robot, and then use the question-and-answer method provided by the embodiment of the present application. , according to the question information input by the user, determine the matching answer, and then reply the answer to the user, and then interact with the user.

In addition, the method provided by the embodiment of the present application is applied to an electronic device, and the electronic device may include a terminal, and may also include a server.

When the electronic device includes a terminal, the methods provided by the embodiments of the present application are executed by the terminal.

Alternatively, FIG. 1 shows a schematic structural diagram of an implementation environment of an embodiment of the present application. Referring to FIG. 1 , when the electronic device includes a terminal 101 and a server 102, the terminal 101 and the server 102 are connected through a communication network, and when the terminal 101 obtains the first After the question information, the first question information is sent to the server 102 through the communication network, the server 102 determines the target answer information according to the method provided by the embodiment of the present application, and then feeds back the target answer information to the terminal 101, and the user can pass the terminal 101 View the target problem information.

Wherein, the terminal may be a mobile phone, a tablet computer, a computer and other types of terminals, and the server may be a server, a server cluster composed of several servers, or a cloud computing service center.

FIG. 2 is a flowchart of a question and answer method provided by an embodiment of the present application. Referring to FIG. 2 , the method includes:

201. Acquire a first question vector of first question information and a second question vector of at least one piece of second question information.

Wherein, at least one piece of second question information is question information obtained before the first question information.

202. Call the answer generation model, and obtain at least one piece of answer information according to the first question vector and at least one second question vector.

The answer generation model is used to generate answer information matched by any question vector according to any question vector.

203. Search in the question answering database according to the first question information or the first question vector.

204. Invoke the sorting model, sort the obtained multiple pieces of answer information in order of matching degree with the first question information from high to low, and determine the answer information ranked first as the target answer information.

The method provided by the embodiment of the present application adopts the method of automatic generation and retrieval based on the model to obtain the answer information, and then invokes the sorting model to sort the obtained multiple pieces of answer information according to the matching degree between the answer information and the question information, and obtain the ranking. The answer information in the first place is the answer information that best matches the first question information, which ensures that the answer information can be generated even if the answer information cannot be retrieved by the retrieval method, and there will be no missing answer information. , and also comprehensively considers the influence of context and the degree of matching between answer information and question information, which improves the accuracy of answer information.

In one possible implementation, the method further includes:

Obtain the sample question information and the corresponding sample answer information, as well as the sample matching degree between the sample question information and the sample answer information;

According to the sample question information, sample answer information and sample matching degree, the ranking model is trained to obtain the trained ranking model.

In another possible implementation, the sorting model is trained according to the sample question information, the sample answer information and the sample matching degree, and the trained sorting model is obtained, including:

Input the sample question information and sample answer information into the ranking model, and obtain the predicted matching degree between the sample question information and the sample answer information;

According to the sample matching degree and the predicted matching degree, the sorting model is trained, and the trained sorting model is obtained.

In another possible implementation manner, the ranking model includes multiple matching layers, a fusion layer, and a ranking layer, and the ranking model is called, and according to the order of matching degree with the first question information from high to low, the obtained more The answer information is sorted, and the first answer information is determined as the target answer information, including:

Call multiple matching layers to obtain the matching degree of each answer information and the first question information respectively;

Call the fusion layer to obtain the fusion matching degree of multiple matching degrees of each answer information;

The sorting layer is called to sort multiple pieces of answer information according to the order of fusion matching degree from high to low, and the answer information ranked first is determined as the target answer information.

In another possible implementation manner, acquiring the first question vector of the first question information includes:

Perform word segmentation processing on the first question information to obtain a plurality of first words after word segmentation;

Obtain similar words of multiple first words;

Each time at least one first word is replaced with a corresponding similar word to generate a piece of similar question information;

Obtain a first question vector according to the first question information and at least one piece of similar question information.

In another possible implementation manner, a plurality of similar words of the first word are obtained, including:

From the knowledge database, obtain similar words associated with each of the first words in the plurality of first words, and the knowledge database is used to store the similar words associated with each word and each word; or,

Obtain the similarity between each first word in the plurality of first words and at least one preset word, and use a preset word whose similarity with any first word is greater than the first preset similarity as any first word similar words.

In another possible implementation manner, obtaining the first question vector according to the first question information and at least one piece of similar question information, including:

The average vector of the feature vector of at least one piece of similar question information and the feature vector of the first question information is obtained as the first question vector.

In another possible implementation manner, the answer generation model is called to obtain at least one piece of answer information according to the first question vector and at least one second question vector, including:

splicing the first problem vector and at least one second problem vector to obtain a third problem vector;

The answer generation model is called to obtain at least one piece of answer information according to the third question vector.

In another possible implementation manner, searching in the question answering database according to the first question information or the first question vector, including:

splicing the first problem vector and at least one second problem vector to obtain a third problem vector;

Obtain at least one piece of answer information according to the similarity between the third question vector and the feature vector of each preset answer information in the question answering database, and the similarity between the feature vector of at least one piece of answer information and the third question vector is greater than that of other preset answers The similarity between the feature vector of the answer information and the third question vector.

In another possible implementation manner, the at least one second question vector includes multiple second question vectors, and the answer generation model is invoked to obtain at least one piece of answer information according to the first question vector and the at least one second question vector, including:

Perform a weighted average on multiple second problem vectors to obtain a fourth problem vector;

Splicing the first problem vector and the fourth problem vector to obtain a fusion problem vector;

According to the fused question vector, the answer generation model is called to obtain at least one piece of answer information.

In another possible implementation manner, searching in the question answering database according to the first question information or the first question vector, including:

Perform a weighted average on a plurality of second problem vectors to obtain a fourth problem vector;

Splicing the first problem vector and the fourth problem vector to obtain a fusion problem vector;

Obtain at least one piece of answer information according to the similarity between the fusion question vector and the feature vector of each preset answer information in the question answering database, and the similarity between the feature vector of at least one piece of answer information and the fusion question vector is greater than that of other preset answer information The similarity between the feature vector of and the fusion problem vector.

In another possible implementation manner, searching in the question answering database according to the first question information or the first question vector, including:

According to the first question vector, perform retrieval in the question answering database to determine at least one fifth question vector, and at least one fifth question vector includes question vectors similar to the first question vector; or,

According to the first question vector and the at least one second question vector, perform a search in the question answering database to determine at least one fifth question vector, and the at least one fifth question vector includes a question vector similar to the first question vector or at least one second question vector at least one of similar problem vectors;

Obtain at least one piece of answer information of the fifth question vector from the question answering database as at least one piece of answer information.

In another possible implementation manner, searching in the question answering database according to the first question information or the first question vector, including:

Obtain at least one piece of answer information according to the similarity between the first question vector and the feature vector of each preset answer information in the question answering database, and the similarity between the feature vector of at least one piece of answer information and the first question vector is greater than that of other preset answers The similarity between the feature vector of the answer information and the fused question vector.

In another possible implementation manner, searching in the question answering database according to the first question information or the first question vector, including:

Perform word segmentation processing on the first question information to obtain keywords or entities of the first question information;

Perform word segmentation on the preset answer information in the question-and-answer database to obtain keywords or entities of the preset answer information;

At least one piece of answer information is acquired according to the keywords or entities of the first question information and the keywords or entities of the preset answer information in the question-and-answer database.

FIG. 3 is a flowchart of a question and answer method provided by an embodiment of the present application. Referring to FIG. 3 , the method is applied to an electronic device, and the method includes:

301. Acquire a first question vector of the first question information and a second question vector of at least one piece of second question information.

The first question information is the question information currently input by the user. For example, the first question information may be question information for a user to inquire about an operation, or for a user to inquire about a name, or for a user to inquire about the price of an item, and so on.

In the embodiment of the present application, when answering the first question information, the answer is not only based on the obtained first question vector of the first question information, but also needs to obtain at least one piece of second question information before the first question information, Subsequently, the answer is made according to the first question information and at least one second question information.

The first question information is the question currently asked by the user. Before the first question information, the user has also asked other questions. Therefore, at least one second question information before the first question information is acquired, and the user's Answering the first question information of the user based on the previously asked question information can improve the accuracy of the generated answer information.

In a possible implementation manner, the obtained at least one piece of second question information is input into a vector coding model, and a second question vector of the at least one second question information is obtained. Among them, the vector encoding model is used to obtain the question vector of any question information.

Optionally, perform word segmentation processing on the first question information, obtain a plurality of first words after word segmentation, obtain similar words of the plurality of first words, replace at least one first word with a corresponding similar word each time, and generate a Similar problem information, obtain a first problem vector according to the first problem information and at least one piece of similar problem information.

Among them, similar words of the first word are obtained, and after the first word is replaced with similar words, similar question information similar to the first question information can be obtained. The expansion of the first question information can obtain more other question information similar to the first question information, and can fully consider different expressions of the same information.

After obtaining the similar words of multiple first words of the segmented words, at least one first word can be replaced with a corresponding similar word, for example, one of the first words is replaced with a corresponding similar word, and two of the first words are replaced with corresponding similar words. A word is replaced with a corresponding similar word, etc., and then a replaced piece of similar question information can be generated, and then a first question vector can be obtained according to the first question information and at least one piece of similar question information.

In a possible implementation manner, similar words associated with each of the multiple first words are obtained from the knowledge database.

Wherein, the knowledge database is used to store similar words associated with each word and each word. And the knowledge database can describe the relationship between each word, establish an association between each word, and form a network structure between the words.

When word segmentation is performed on the first question information, and then the knowledge database is queried according to the first word obtained from the word segmentation, the similar words associated with the first word in the knowledge database can be determined.

In another possible implementation manner, the similarity between each first word in the plurality of first words and at least one preset word is obtained, and the similarity with any first word is greater than the first preset similarity. Let words be similar to any of the first words.

In the embodiment of the present application, after word segmentation is performed on the first question information, a plurality of first words are obtained, and then the similarity between each first word and at least one preset word is calculated, and then the similarity between each first word and at least one preset word is calculated. Whether the similarity of at least one preset word is greater than the first preset similarity, when it is determined that the similarity between any first word and the preset word is greater than the first preset similarity, then the preset word is used as the any one Similar words for the first word.

Wherein, the first preset similarity is set by a server, or set by a technician, or set by other methods. The first preset similarity may be 0.7, 0.8 or other values.

For example, when the first question information input by the user is "I want to inquire about this order", the multiple first words obtained by word segmentation are "I", "Want", "Inquiry", "Ask", "This" , "order", it can be determined that the similar word of "query" is "find", and the similar word of "order" is "billing", replace "query" with "find", the generated similar question information is "I want to find Check this order", replace "order" with "billing", the generated similar question information is "I want to check this bill", and at the same time, replace "query" with "find", and replace "order" with " Bill", the generated similar question information is "I want to look up this bill", and then the first question vector is obtained according to the first question information and at least one similar question information.

Optionally, an average vector of the feature vector of at least one piece of similar question information and the feature vector of the first question information is obtained as the first question vector.

When generating the first question vector according to the first question information and at least one piece of similar question information, first obtain the feature vector of the at least one similar question information and the feature vector of the first question information, and then calculate the feature of the at least one similar question information The average vector of the vector and the eigenvectors of the first question information, as the first question vector.

In a possible implementation manner, after obtaining the first question information and at least one similar question information, input the first question information and at least one similar question information into the sentence vector coding model, and then obtain the first question information Eigenvectors and eigenvectors of similar question information.

302. Call the answer generation model, and obtain at least one piece of answer information according to the first question vector and at least one second question vector.

Wherein, the answer generation model is used to generate answer information matched by any question vector according to any question vector.

Among them, the answer generation model can be a seq2seq-attention (a model structure) model, an encoder-decoder (a model structure) model, a transformer (a model structure) model, and so on.

Optionally, the answer generation model includes a vector decoding module, an attention mechanism weighting module, an answer error correction module and a generative answer module. For example, as shown in Figure 4, after the question vector is input into the answer generation model, The question vector is weighted by the attention mechanism weighting module to obtain the most influential part in the question vector, and then the predicted answer is obtained through the vector decoding module, and the predicted answer is corrected by the answer error correction module. Finally, the answer information is obtained, which is also the generative answer information.

After the first question vector and the at least one second question vector are obtained, the answer generation model is invoked, and at least one piece of answer information can be obtained according to the first question vector and the at least one second question vector.

In a possible implementation manner, the first question vector and at least one second question vector are spliced to obtain a third question vector, the answer generation model is called, and at least one piece of answer information is obtained according to the third question vector.

The first question vector and at least one second question vector are spliced in sequence to obtain a spliced third question vector, and then the third question vector is input into the answer generation model to generate at least one piece of answer information.

Optionally, when splicing the first problem vector and the at least one second problem vector in sequence, the splicing is performed in the order in which the first problem vector and the at least one second problem vector are obtained, and a third problem vector after splicing is obtained. .

Alternatively, in a random order, the first problem vector and at least one second problem vector are spliced in sequence to obtain a spliced third problem vector.

In another possible implementation manner, the obtained at least one second problem vector includes multiple second problem vectors.

In order to reduce the data amount of the second problem vector, firstly weighted and average multiple second problem vectors to obtain the fourth problem vector, and then splicing the first problem vector and the fourth problem vector to obtain the fusion problem vector, according to the fusion problem Vector, call the answer generation model to obtain at least one answer information.

Optionally, the called answer generation model can directly process the fused question vector to generate a piece of answer information.

Alternatively, the called answer generation model can use multiple processing layers to process the fused question vector respectively to obtain multiple processed question vectors, and then each processing layer will generate an answer information according to a question vector, then finally Multiple answer information can be obtained.

Optionally, in the process of performing the weighted average on the plurality of second problem vectors, the weight of each second problem vector is determined according to the sequence of the plurality of second problem vectors, and then according to the weight of each second problem vector. A weighted average is performed on a plurality of second problem vectors to obtain a fourth problem vector.

For example, the reciprocal of the interval between each second problem vector and the first problem vector is used as the weight of each second problem vector. When the interval between the second problem vector and the first problem vector is 1, the weight is 1. When the interval between the second problem vector and the first problem vector is 2, the weight is 1/2, and so on. The larger the interval between the second problem vector and the first problem vector, the weight smaller.

Among them, the process from obtaining question information, obtaining answer information according to the question information, and outputting the answer information is one round of questioning and answering. Therefore, the interval between the second question vector and the first question vector is the number of rounds between the first question vector and the first question vector. The number of rounds of the question answering process between the second question vectors.

In addition, before calling the answer generation model, the answer generation model needs to be trained first, and then the trained answer generation model is called. The step of training the answer generation model includes: first obtaining an initial answer generation model or an answer generation model that has been trained one or more times, and then obtaining a sample question vector and corresponding sample answer information, according to the sample question vector and corresponding The sample answer information is used to train the initial answer generation model to obtain the trained answer generation model.

In the training process, input at least one sample question vector into the answer generation model, obtain the predicted answer information based on the answer generation model, and then obtain the error between the sample answer information and the predicted answer information, and adjust the answer generation model to The error obtained by the adjusted answer generation model is converged, and the training of the answer generation model is completed.

303. Search in the question answering database according to the first question information or the first question vector.

Among them, there are multiple pieces of answer information stored in the question and answer database, and each piece of answer information in the multiple pieces of answer information corresponds to question information. Therefore, according to the obtained first question vector and at least one second question vector, in the question and answer The database is searched, and after the search, it can be determined whether there is answer information that matches the first question information.

During the retrieval process in the question-and-answer database, retrieval may be performed according to the first question information, and may also be retrieved according to the first question vector, and the answer information may be obtained through subsequent retrieval.

In a possible implementation manner, the first question vector and at least one second question vector are spliced together to obtain a third question vector. According to the relationship between the third question vector and the feature vector of each preset answer information in the question answering database The similarity of at least one piece of answer information is obtained, and the similarity between the feature vector of at least one piece of answer information and the third question vector is greater than the similarity between the feature vectors of other preset answer information and the third question vector.

Among them, the feature vector of each preset answer information in the question and answer database can be obtained, and then the similarity between the third question vector and the feature vector of each preset answer information can be calculated, and then the similarity is from large to small. Sort the preset answer information in order, and obtain at least one piece of answer information according to the sorting order.

Optionally, according to the arrangement order, the preset answer information of the first preset number is selected as the answer information obtained by retrieval.

Wherein, the preset number can be set by the server, or set by the technician, or set by other methods. The preset number can be 2, 3, 4 or other values.

Optionally, according to the similarity between the acquired third question vector and the feature vector of each preset answer information, the preset answer information whose similarity is greater than the preset similarity is selected as the retrieved answer information.

In another possible implementation manner, a weighted average is performed on a plurality of second question vectors to obtain a fourth question vector, the first question vector and the fourth question vector are spliced to obtain a fusion question vector, and according to the fusion question vector and the question and answer The similarity between the feature vectors of each preset answer information in the database, obtain at least one piece of answer information, and the similarity between the feature vector of at least one piece of answer information and the fusion question vector is greater than that of other preset answer information. similarity of vectors.

The process of performing the weighted average on the plurality of second problem vectors is similar to the weighted average process in the above steps, and details are not repeated here. In addition, the process of calculating the similarity between the fusion feature vector and the feature vector of the preset answer information is similar to the process of calculating the similarity between the third feature vector and the feature vector of the preset answer information in the above process. No longer.

In another possible implementation manner, at least one piece of answer information is acquired according to the similarity between the first question vector and the feature vector of each preset answer information in the question answering database.

Wherein, the similarity between the feature vector of the at least one piece of answer information and the first question vector is greater than the similarity between the feature vector of other preset answer information and the first question vector.

The process of acquiring at least one piece of answer information according to the similarity is similar to the above-mentioned process of acquiring at least one piece of answer information, and details are not repeated here.

Optionally, in the process of obtaining at least one piece of answer information according to the first question vector, the quantity of the obtained answer information can be determined by setting a preset quantity. For example, when the preset quantity is 1, the obtained answer information The information is 1, and when the preset number is set to 2, the obtained answer information is multiple.

Or, when the answer information is obtained according to the preset similarity, since the similarity between the obtained third question vector or the fusion feature vector and the feature vector of the preset answer information is not fixed, if both are smaller than the preset similarity, the similarity is not fixed. Answer information will be obtained. If only one is greater than the preset similarity, only one answer information will be obtained. If there are multiple similarities greater than the preset similarity, multiple answer information will be obtained.

In another possible implementation manner, according to the first question vector, a question-and-answer database is searched to determine at least one fifth question vector. Wherein, at least one fifth question vector includes a question vector similar to the first question vector.

Optionally, the similarity between at least one fifth question vector and the first question vector is greater than the similarity between other preset question vectors and the first question vector.

In another possible implementation manner, according to the first question vector and at least one second question vector, retrieval is performed in the question answering database, and at least one fifth question vector is determined, and the at least one fifth question vector includes a similarity between the first question vector and the At least one of the question vector or at least one question vector similar to the second question vector is obtained from the question answering database, and the answer information of at least one fifth question vector is obtained as the at least one piece of answer information obtained by retrieval.

Among them, in the process of obtaining at least one fifth problem vector through retrieval, since a problem vector similar to the problem vector is determined, in order to reduce the situation that the fifth problem vector cannot be obtained, the obtained fifth problem vector can be the same as the first problem vector. A vector similar to the problem vector, or a vector similar to the second problem vector.

Optionally, since the fifth problem vector is obtained according to the first problem vector and at least one second problem vector, there will be three situations: the fifth problem vector is not obtained, a fifth problem vector is obtained, A plurality of fifth question vectors are obtained.

Optionally, the first question vector and at least one second question vector are spliced to obtain a third question vector, and according to the third question vector, the question and answer database is searched to determine at least one fifth question vector.

Among them, at least one fifth question vector is obtained according to the similarity between the third question vector and each preset question vector in the question answering database, and the similarity between at least one fourth question vector and the third question vector is greater than other preset question vectors The similarity of the question vector to the third question vector.

Optionally, a weighted average is performed on a plurality of second question vectors to obtain a sixth question vector, the first question vector and the sixth question vector are spliced to obtain a fusion question vector, and a search is performed in the question answering database according to the fusion question vector. , determine at least one fifth problem vector.

Among them, at least one fifth question vector is obtained according to the similarity between the fusion question vector and each preset question vector in the question answering database, and the similarity between at least one fifth question vector and the third question vector is greater than that of other preset questions The similarity of the vector to the fusion problem vector.

It should be noted that, the embodiments of the present application are only described by taking an example of performing retrieval in a question-and-answer database to obtain answer information according to the first question vector. In another embodiment, at least one piece of answer information can be obtained by searching in the question and answer database directly according to the first question information.

In a possible implementation manner, word segmentation processing is performed on the first question information, keywords or entities in the first question information are obtained, and word segmentation processing is performed on the preset answer information in the question-and-answer database to obtain the information of the preset answer information. The keyword or entity, when the keyword or entity in the first question information matches the keyword or entity in the preset answer information, the preset answer information is determined as the answer information.

In another possible implementation manner, word segmentation processing is performed on the first question information, keywords or entities of the first question information are obtained, and word segmentation processing is performed on the preset question information in the question-and-answer database to obtain the key of the preset answer information. word or entity, according to the keywords or entities of the first question information and the keywords or entities of the preset answer information in the question and answer database, search in the question and answer database, and determine at least one third question information similar to the first question information , and then determine the answer information corresponding to at least one piece of third question information as the answer information.

For example, as shown in Figure 5, during the retrieval process, multiple preset answer information is obtained from the question-and-answer database, and the feature vector of each preset answer information is obtained according to the vector coding model, and then the preset answer information and the current The similarity of the question vector, and the answer information is obtained according to the obtained similarity. Alternatively, the keywords or entities of the preset answer information and the keywords or entities of the current question information can also be obtained, and the similarity between the preset answer information and the keywords and the keywords of the current question information can be calculated, or the preset answer information can be calculated. The similarity between the answer information and the entity and the entity of the current question information, and the answer information is obtained according to the obtained similarity. The answer information obtained by the above two methods can be regarded as retrieval type answer information.

304. Invoke multiple matching layers in the sorting model to obtain the matching degree of each piece of answer information and the first question information respectively.

Among them, the ranking model includes multiple matching layers, a fusion layer and a ranking layer. Each matching layer in the multiple matching layers is used to obtain the matching degree of the question information and the answer information. The fusion layer is used to fuse the matching degree of each question information and answer information to obtain the fusion matching degree of the question information and the answer information. The sorting layer is used to sort the answer information according to the fusion matching degree of the question information and each piece of answer information, and take the answer information ranked first as the target answer information of the question information.

After obtaining multiple pieces of answer information, multiple matching layers in the sorting model are called, and a matching degree between each answer information and the first question information can be obtained through one matching layer, then the same answer information and the first question information pass through multiple matching layers. The matching layer can obtain multiple matching degrees.

Optionally, the matching degrees of the first question information and the answer information are acquired in different ways in multiple matching layers.

Among them, LSTM (Long short-term memory, long short-term memory), CNN (Convolutional Neural Networks, convolutional neural network), MLP (Multi-Layer Perceptron, multi-layer neural network), GBDT (Gradient Boosting Decision Tree, gradient) can be used Machine learning algorithms such as improving iterative decision tree) to obtain the matching degree, or use other methods to obtain the matching degree.

305. Call the fusion layer in the sorting model to obtain a fusion matching degree of multiple matching degrees of each piece of answer information.

After obtaining multiple matching degrees of each answer information, the fusion layer is then called to obtain the fusion matching degree of multiple matching degrees of each answer information.

Optionally, the average value of multiple matching degrees of each piece of answer information is obtained as the fusion matching degree of the answer information.

Optionally, each matching layer has a corresponding weight, and the sum of the corresponding weights of the multiple matching layers is 1, then according to the corresponding weight of each matching layer, the multiple matching degrees of each piece of answer information are weighted. Averaged to obtain the fusion matching degree of the answer information.

306. Call the sorting layer in the sorting model, sort the multiple pieces of answer information in the order of the fusion matching degree from high to low, and determine the answer information ranked first as the target answer information.

For example, as shown in FIG. 6 , after obtaining multiple pieces of answer information, the sorting model is invoked to determine the target answer information.

In addition, in this embodiment of the present application, when the sorting model is invoked to obtain target answer information from multiple pieces of answer information, the priority of at least one piece of answer information obtained through step 303 is higher than the priority of at least one piece of answer information obtained through step 302 Therefore, when the sorting model sorts multiple pieces of answer information, the weight set for the answer information obtained in step 303 is greater than the weight set for the answer information obtained in step 303, then when sorting according to the fusion matching degree, first Through the set weights, the fusion matching degree is weighted to obtain the weighted fusion matching degree, and then sorting is performed according to the weighted fusion matching degree to obtain the target answer information.

It should be noted that the embodiment of the present application only takes the sorting model including multiple matching layers, one fusion layer and one sorting layer as an example for description. In another embodiment, steps 304 to 306 are optional steps, and in There is no need to set up multiple matching layers in the ranking model, and the ranking model can be directly called to obtain a matching degree between the answer information and the first question information, and determine the target answer information according to the obtained matching degree.

In a possible implementation manner, the sorting model is invoked to sort at most pieces of answer information in descending order of matching degree with the first question information, and the answer information ranked first is determined as the target answer information .

In the embodiment of the present application, the sorting model is invoked to obtain the matching degree of each piece of answer information among the multiple pieces of answer information and the first question information. Sort multiple pieces of answer information in the order of the matching degree from high to low, and determine the answer information in the first place as the target answer information.

In addition, before calling the sorting model, it is necessary to train the sorting model to obtain the trained sorting model, and then call the trained sorting model to obtain the target answer. The process of training a ranking model is as follows:

Obtain the sample question information and the corresponding sample answer information, as well as the sample matching degree between the sample question information and the sample answer information, train the sorting model according to the sample question information, the sample answer information and the sample matching degree, and obtain the trained sorting model.

The sample question information and the sample answer information are in a corresponding relationship, that is, the answer of the sample question information is the sample answer information. In addition, the sample matching degree of the sample question information and the sample answer information is used to indicate that the sample question information and the sample answer information match.

After obtaining the sample question information and the corresponding sample answer information, and the sample matching degree between the sample question information and the corresponding sample answer information, the sorting model can be trained according to the sample question information, the sample answer information and the sample matching degree. Get the trained ranking model.

Optionally, input the sample question information and sample answer information into the ranking model, obtain the predicted matching degree between the sample question information and the sample answer information, train the ranking model according to the sample matching degree and the predicted matching degree, and obtain the training results. Sort the model.

In the training process, input at least one sample question information and corresponding sample answer information into the ranking model, obtain the predicted matching degree based on the ranking model, and then obtain the error between the sample matching degree and the predicted matching degree, and perform the ranking model. Adjust so that the error obtained by the adjusted ranking model converges, and the training of the ranking model is completed.

The flow of the method provided by this embodiment of the present application is shown in FIG. 7 , step 301 is executed by the input module, step 302 is executed by the answer generation module to obtain at least one generative answer, and step 303 is executed by the retrieval answer module to obtain at least one retrieval formula Answers, steps 304-306 are executed by the answer sorting module, which sorts at least one generative answer and at least one retrieval answer, and obtains the answer of the current round.

In addition, the process in the input module is shown in Figure 8. First, the first question information input by the user in the current round is obtained, the first question information is word-segmented, and then the words after word segmentation are expanded to obtain similar question information. The first question information, similar question information and second question information, the second question information is the scene context information, call the scene context coding model to obtain the scene context information vector, and call the sentence vector coding model to obtain the first question information and similar questions The first question vector of the information, and then generate a fusion feature vector according to the scene context information vector and the first question vector. The fusion feature vector is the input of the subsequent generative module and retrieval module, and becomes the scene context information of the next question information. .

All the above-mentioned optional technical solutions can be combined arbitrarily to form optional embodiments of the present disclosure, which will not be repeated here.

The method provided by the embodiment of the present application adopts the method of automatic generation and retrieval based on the model to obtain the answer information, and then invokes the sorting model to sort the obtained multiple pieces of answer information according to the matching degree between the answer information and the question information, and obtain the ranking. The answer information in the first place is the answer information that best matches the first question information, which ensures that the answer information can be generated even if the answer information cannot be retrieved by the retrieval method, and there will be no missing answer information. , and also comprehensively considers the influence of context and the degree of matching between answer information and question information, which improves the accuracy of answer information.

In addition, multiple matching degrees of question information and answer information are obtained by adopting a variety of ways to obtain matching degrees, and the multiple matching degrees are fused to obtain a fusion matching degree, which integrates multiple matching degrees and is more accurate. Therefore, determining the answer information with the highest fusion matching degree as the target answer information can improve the accuracy of the target answer information.

In addition, by adopting the method of word segmentation and then expanding the words, the expansion of the first question information is realized, and more question information similar to the first question information can be obtained, and the different expressions of the same information can be fully considered to improve the subsequent acquisition. accuracy of the answer information.

FIG. 9 is a schematic structural diagram of a question and answer device provided by an embodiment of the present application. Referring to Figure 9, the device includes:

Obtaining module 901, configured to obtain a first question vector of first question information and a second question vector of at least one piece of second question information, where the at least one piece of second question information is question information obtained before the first question information ;

The generation module 902 is configured to call an answer generation model, obtain at least one piece of answer information according to the first question vector and at least one second question vector, and the answer generation model is used to generate the any question according to any question vector vector matching answer information;

A retrieval module 903, configured to perform retrieval in the question and answer database according to the first question information or the first question vector;

The sorting module 904 is used to call the sorting model, sort the obtained multiple pieces of answer information according to the order of matching degree with the first question information from high to low, and determine the first answer information as The target answer information.

The device provided by the embodiment of the present application adopts the method of automatic generation and retrieval based on the model to obtain the answer information, and then invokes the sorting model, sorts the obtained multiple pieces of answer information according to the matching degree between the answer information and the question information, and obtains the ranking. The answer information in the first place is the answer information that best matches the first question information, which ensures that the answer information can be generated even if the answer information cannot be retrieved by the retrieval method, and there will be no missing answer information. , and also comprehensively considers the influence of context and the degree of matching between answer information and question information, which improves the accuracy of answer information.

In a possible implementation, the apparatus further includes:

The obtaining module 901 is configured to obtain sample question information and corresponding sample answer information, and a sample matching degree between the sample question information and the sample answer information;

The training module 905 is configured to train the ranking model according to the sample question information, the sample answer information and the sample matching degree to obtain a trained ranking model.

In another possible implementation, the training module 905 includes:

an input unit 9051, configured to input the sample question information and the sample answer information into the ranking model, and obtain the predicted matching degree of the sample question information and the sample answer information;

A training unit 9052, configured to train the ranking model according to the sample matching degree and the predicted matching degree, to obtain a trained ranking model.

In another possible implementation, the ranking model includes multiple matching layers, a fusion layer and a ranking layer, and the generation module 902 includes:

a matching degree obtaining unit 9021, configured to invoke the multiple matching layers to obtain the matching degree of each piece of answer information and the first question information respectively;

The fusion unit 9022 is used to call the fusion layer to obtain the fusion matching degree of multiple matching degrees of each piece of answer information;

The sorting unit 9023 is configured to call the sorting layer, sort the multiple pieces of answer information in the order of the fusion matching degree from high to low, and determine the answer information ranked first as the target answer information.

In another possible implementation manner, the obtaining module 901 is configured to execute any one of the following:

Perform word segmentation processing on the first question information to obtain a plurality of first words after word segmentation;

obtaining similar words of the plurality of first words;

Each time at least one first word is replaced with a corresponding similar word to generate a piece of similar question information;

Obtain the first question vector according to the first question information and the at least one piece of similar question information.

In another possible implementation manner, the obtaining module 901 is configured to execute any one of the following:

Obtain, from a knowledge database, similar words associated with each of the plurality of first words, where the knowledge database is used to store similar words associated with each word; or,

Obtain the similarity between each first word in the plurality of first words and at least one preset word, and use a preset word whose similarity with any first word is greater than the first preset similarity as the arbitrary first word. A similar word to the first word.

In another possible implementation manner, the obtaining module 901 is configured to obtain the average vector of the feature vector of the at least one piece of similar question information and the feature vector of the first question information, as the first question vector.

In another possible implementation manner, the generating module 902 is configured to perform any one of the following:

splicing the first problem vector and the at least one second problem vector to obtain a third problem vector;

The answer generation model is called, and the at least one piece of answer information is acquired according to the third question vector.

In another possible implementation manner, the retrieval module 903 is configured to perform any one of the following:

splicing the first problem vector and the at least one second problem vector to obtain a third problem vector;

According to the similarity between the third question vector and the feature vector of each preset answer information in the question answering database, at least one piece of answer information is obtained, and the feature vector of the at least one piece of answer information is related to the third question The similarity of the vectors is greater than the similarity between the feature vectors of other preset answer information and the third question vector.

In another possible implementation manner, the at least one second problem vector includes a plurality of second problem vectors, and the generating module 902 is configured to perform any one of the following:

performing a weighted average on the plurality of second problem vectors to obtain a fourth problem vector;

Splicing the first problem vector and the fourth problem vector to obtain a fusion problem vector;

According to the fusion question vector, the answer generation model is called to obtain the at least one piece of answer information.

In another possible implementation manner, the retrieval module 903 is configured to perform any one of the following:

performing a weighted average on the plurality of second problem vectors to obtain a fourth problem vector;

Splicing the first problem vector and the fourth problem vector to obtain a fusion problem vector;

At least one piece of answer information is obtained according to the similarity between the fusion question vector and the feature vector of each preset answer information in the question answering database, and the feature vector of the at least one piece of answer information is the difference between the fusion question vector and the fusion question vector. The similarity is greater than the similarity between the feature vector of other preset answer information and the fusion question vector.

In another possible implementation manner, the retrieval module 903 is configured to perform any one of the following:

According to the first question vector, perform retrieval in the question answering database to determine at least one fifth question vector, where the at least one fifth question vector includes question vectors similar to the first question vector; or,

According to the first question vector and the at least one second question vector, searching in the question answering database to determine at least one fourth question vector, the at least one fourth question vector including the first question vector is similar At least one of the problem vectors of or the problem vectors that are similar to the at least one second problem vector;

From the question and answer database, the answer information of the at least one fourth question vector is obtained as the at least one piece of answer information.

In another possible implementation manner, the retrieval module 903 is configured to perform any one of the following:

Perform word segmentation processing on the first question information to obtain keywords or entities of the first question information;

Perform word segmentation processing on the preset answer information in the question and answer database to obtain keywords or entities of the preset answer information;

At least one piece of answer information is acquired according to the keywords or entities of the first question information and the keywords or entities of the preset answer information in the question-and-answer database.

It should be noted that when the question-and-answer device provided in the above-mentioned embodiments conducts questions and answers, only the division of the above-mentioned functional modules is used for illustration. The internal structure of the device is divided into different functional modules to complete all or part of the functions described above. In addition, the question-answering device and question-answering method embodiments provided by the above embodiments belong to the same concept, and the specific implementation process thereof is detailed in the method embodiments, which will not be repeated here.

FIG. 11 is a schematic structural diagram of a terminal provided by an embodiment of the present application. The terminal 1100 may be a portable mobile terminal, such as: a smart phone, a tablet computer, an MP3 player (Moving Picture Experts Group AudioLayer III, Moving Picture Experts Group Audio Layer III), MP4 (Moving Picture Experts Group Audio Layer IV, Moving Picture Experts Group Audio Layer IV) Compression standard audio layer 4) Player, notebook computer, desktop computer, head-mounted device, or any other intelligent terminal. Terminal 1100 may also be called user equipment, portable terminal, laptop terminal, desktop terminal, and the like by other names.

Generally, the terminal 1100 includes: a processor 1101 and a memory 1102 .

The processor 1101 may include one or more processing cores, such as a 4-core processor, an 8-core processor, and the like. The processor 1101 may use at least one hardware form of DSP (Digital Signal Processing, digital signal processing), FPGA (Field-Programmable Gate Array, field programmable gate array), and PLA (Programmable Logic Array, programmable logic array). accomplish. The processor 1101 may also include a main processor and a coprocessor. The main processor is a processor used to process data in a wake-up state, also called a CPU (Central Processing Unit, central processing unit); A low-power processor for processing data in a standby state. In some embodiments, the processor 1101 may be integrated with a GPU (Graphics Processing Unit, image processor), and the GPU is used for rendering and drawing the content that needs to be displayed on the display screen. In some embodiments, the processor 1101 may further include an AI (Artificial Intelligence, artificial intelligence) processor, where the AI processor is used to process computing operations related to machine learning.

Memory 1102 may include one or more computer-readable storage media, which may be non-transitory. Memory 1102 may also include high-speed random access memory, as well as non-volatile memory, such as one or more disk storage devices, flash storage devices. In some embodiments, the non-transitory computer-readable storage medium in the memory 1102 is used to store at least one instruction, the at least one instruction is used by the processor 1101 to implement the question answering method provided by the method embodiments of the present application .

In some embodiments, the terminal 1100 may optionally further include: a peripheral device interface 1103 and at least one peripheral device. The processor 1101, the memory 1102 and the peripheral device interface 1103 may be connected through a bus or a signal line. Each peripheral device can be connected to the peripheral device interface 1103 through a bus, a signal line or a circuit board. Specifically, the peripheral device includes: at least one of a radio frequency circuit 1104 , a display screen 1105 , a camera assembly 1106 , an audio circuit 1107 , a positioning assembly 1108 and a power supply 11011 .

The peripheral device interface 1103 may be used to connect at least one peripheral device related to I/O (Input/Output) to the processor 1101 and the memory 1102 . In some embodiments, processor 1101, memory 1102, and peripherals interface 1103 are integrated on the same chip or circuit board; in some other embodiments, any one of processor 1101, memory 1102, and peripherals interface 1103 or The two can be implemented on a separate chip or circuit board, which is not limited in this embodiment.

The radio frequency circuit 1104 is used for receiving and transmitting RF (Radio Frequency, radio frequency) signals, also called electromagnetic signals. The radio frequency circuit 1104 communicates with communication networks and other communication devices via electromagnetic signals. The radio frequency circuit 1104 converts electrical signals into electromagnetic signals for transmission, or converts received electromagnetic signals into electrical signals. Optionally, the radio frequency circuit 1104 includes an antenna system, an RF transceiver, one or more amplifiers, a tuner, an oscillator, a digital signal processor, a codec chipset, a subscriber identity module card, and the like. The radio frequency circuit 1104 may communicate with other terminals through at least one wireless communication protocol. The wireless communication protocols include, but are not limited to, metropolitan area networks, mobile communication networks of various generations (2G, 3G, 4G and 8G), wireless local area networks and/or WiFi (Wireless Fidelity, wireless fidelity) networks. In some embodiments, the radio frequency circuit 1104 may further include a circuit related to NFC (Near Field Communication, short-range wireless communication), which is not limited in this application.

The display screen 1105 is used for displaying UI (User Interface, user interface). The UI can include graphics, text, icons, video, and any combination thereof. When the display screen 1105 is a touch display screen, the display screen 1105 also has the ability to acquire touch signals on or above the surface of the display screen 1105 . The touch signal can be input to the processor 1101 as a control signal for processing. At this time, the display screen 1105 may also be used to provide virtual buttons and/or virtual keyboards, also referred to as soft buttons and/or soft keyboards. In some embodiments, there may be one display screen 1105, which is arranged on the front panel of the terminal 1100; in other embodiments, there may be at least two display screens 1105, which are respectively arranged on different surfaces of the terminal 1100 or in a folded design; In other embodiments, the display screen 1105 may be a flexible display screen, which is disposed on a curved surface or a folding surface of the terminal 1100 . Even, the display screen 1105 can also be set as a non-rectangular irregular figure, that is, a special-shaped screen. The display screen 1105 can be made of materials such as LCD (Liquid Crystal Display, liquid crystal display), OLED (Organic Light-Emitting Diode, organic light emitting diode).

The camera assembly 1106 is used to capture images or video. Optionally, the camera assembly 1106 includes a front camera and a rear camera. Usually, the front camera is arranged on the front panel of the terminal, and the rear camera is arranged on the back of the terminal. In some embodiments, there are at least two rear cameras, which are any one of a main camera, a depth-of-field camera, a wide-angle camera, and a telephoto camera, so as to realize the fusion of the main camera and the depth-of-field camera to realize the background blur function, the main camera It is integrated with the wide-angle camera to achieve panoramic shooting and VR (Virtual Reality, virtual reality) shooting functions or other integrated shooting functions. In some embodiments, the camera assembly 1106 may also include a flash. The flash can be a single color temperature flash or a dual color temperature flash. Dual color temperature flash refers to the combination of warm light flash and cold light flash, which can be used for light compensation under different color temperatures.

Audio circuitry 1107 may include a microphone and speakers. The microphone is used to collect the sound waves of the user and the environment, convert the sound waves into electrical signals, and input them to the processor 1101 for processing, or to the radio frequency circuit 1104 to realize voice communication. For the purpose of stereo collection or noise reduction, there may be multiple microphones, which are respectively disposed in different parts of the terminal 1100 . The microphone may also be an array microphone or an omnidirectional collection microphone. The speaker is used to convert the electrical signal from the processor 1101 or the radio frequency circuit 1104 into sound waves. The loudspeaker can be a traditional thin-film loudspeaker or a piezoelectric ceramic loudspeaker. When the speaker is a piezoelectric ceramic speaker, it can not only convert electrical signals into sound waves audible to humans, but also convert electrical signals into sound waves inaudible to humans for distance measurement and other purposes. In some embodiments, audio circuitry 1107 may also include a headphone jack.

The positioning component 1108 is used to locate the current geographic location of the terminal 1100 to implement navigation or LBS (Location Based Service, location-based service). The positioning component 1108 may be a positioning component based on the GPS (Global Positioning System, global positioning system) of the United States, the Beidou system of China, the Grenas system of Russia, or the Galileo system of the European Union.

The power supply 1109 is used to power various components in the terminal 1100 . The power source 1109 may be alternating current, direct current, disposable batteries, or rechargeable batteries. When the power source 1109 includes a rechargeable battery, the rechargeable battery can support wired charging or wireless charging. The rechargeable battery can also be used to support fast charging technology.

In some embodiments, the terminal 1100 also includes one or more sensors 1110 . The one or more sensors 1110 include, but are not limited to, an acceleration sensor 1111 , a gyro sensor 1112 , a pressure sensor 1113 , a fingerprint sensor 1114 , an optical sensor 1115 , and a proximity sensor 1116 .

The acceleration sensor 1111 can detect the magnitude of acceleration on the three coordinate axes of the coordinate system established by the terminal 1100 . For example, the acceleration sensor 1111 can be used to detect the components of the gravitational acceleration on the three coordinate axes. The processor 1101 can control the display screen 1105 to display the user interface in a landscape view or a portrait view according to the gravitational acceleration signal collected by the acceleration sensor 1111 . The acceleration sensor 1111 can also be used for game or user movement data collection.

The gyroscope sensor 1112 can detect the body direction and rotation angle of the terminal 1100 , and the gyroscope sensor 1112 can cooperate with the acceleration sensor 1111 to collect 3D actions of the user on the terminal 1100 . The processor 1101 can implement the following functions according to the data collected by the gyro sensor 1112: motion sensing (such as changing the UI according to the user's tilt operation), image stabilization during shooting, game control, and inertial navigation.

The pressure sensor 1113 may be disposed on the side frame of the terminal 1100 and/or the lower layer of the display screen 1105 . When the pressure sensor 1113 is disposed on the side frame of the terminal 1100, the user's holding signal of the terminal 1100 can be detected, and the processor 1101 can perform left and right hand identification or shortcut operations according to the holding signal collected by the pressure sensor 1113. When the pressure sensor 1113 is disposed on the lower layer of the display screen 1105, the processor 1101 controls the operability controls on the UI interface according to the user's pressure operation on the display screen 1105. The operability controls include at least one of button controls, scroll bar controls, icon controls, and menu controls.

The fingerprint sensor 1114 is used to collect the user's fingerprint, and the processor 1101 identifies the user's identity according to the fingerprint collected by the fingerprint sensor 1114, or the fingerprint sensor 1114 identifies the user's identity according to the collected fingerprint. When the user's identity is identified as a trusted identity, the processor 1101 authorizes the user to perform relevant sensitive operations, including unlocking the screen, viewing encrypted information, downloading software, making payments, and changing settings. The fingerprint sensor 1114 may be disposed on the front, back or side of the terminal 1100 . When the terminal 1100 is provided with physical buttons or a manufacturer's logo, the fingerprint sensor 1114 may be integrated with the physical buttons or the manufacturer's logo.

Optical sensor 1115 is used to collect ambient light intensity. In one embodiment, the processor 1101 can control the display brightness of the display screen 1105 according to the ambient light intensity collected by the optical sensor 1115 . Specifically, when the ambient light intensity is high, the display brightness of the display screen 1105 is increased; when the ambient light intensity is low, the display brightness of the display screen 1105 is decreased. In another embodiment, the processor 1101 can also dynamically adjust the shooting parameters of the camera assembly 1106 according to the ambient light intensity collected by the optical sensor 1115 .

A proximity sensor 1116 , also called a distance sensor, is usually disposed on the front panel of the terminal 1100 . The proximity sensor 1116 is used to collect the distance between the user and the front of the terminal 1100 . In one embodiment, when the proximity sensor 1116 detects that the distance between the user and the front of the terminal 1100 is gradually decreasing, the processor 1101 controls the display screen 1105 to switch from the bright screen state to the off screen state; when the proximity sensor 1116 detects When the distance between the user and the front of the terminal 1100 gradually increases, the processor 1101 controls the display screen 1105 to switch from the screen-off state to the screen-on state.

Those skilled in the art can understand that the structure shown in FIG. 11 does not constitute a limitation on the terminal 1100, and may include more or less components than the one shown, or combine some components, or adopt different component arrangements.

12 is a schematic structural diagram of a server provided by an embodiment of the present application. The server 1200 may vary greatly due to different configurations or performance, and may include one or more processors (Central Processing Units, CPU) 1201 and one or more One or more memories 1202, wherein at least one instruction is stored in the memory 1202, and the at least one instruction is loaded and executed by the processor 1201 to implement the methods provided by the above method embodiments. Of course, the server may also have components such as a wired or wireless network interface, a keyboard, and an input/output interface for input and output, and the server may also include other components for implementing device functions, which will not be described here.

The server 1200 may be used to perform the steps performed by the server in the above question and answer method.

An embodiment of the present application further provides an electronic device, the electronic device includes one or more processors and one or more memories, at least one instruction is stored in the one or more memories, and at least one instruction is executed by the one or more processors Load and execute to do what the Q&A method does.

Embodiments of the present application further provide a computer-readable storage medium, where at least one instruction is stored, and the at least one instruction is loaded and executed by a processor to implement operations performed by the question-and-answer method.

Those of ordinary skill in the art can understand that all or part of the steps of implementing the above embodiments can be completed by hardware, or can be completed by instructing relevant hardware through a program, and the program can be stored in a computer-readable storage medium. The storage medium mentioned may be a read-only memory, a magnetic disk or an optical disk, etc.

The above descriptions are only preferred embodiments of the present application, and are not intended to limit the present application. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present application shall be included in the protection of the present application. within the range.

Claims (15)

1. a question and answer method, it is characterised in that the method comprises: acquiring a first question vector of first question information and at least one second question vector of second question information, where the at least one piece of second question information is question information obtained before the first question information; Invoke the answer generation model, obtain at least one piece of answer information according to the first question vector and at least one second question vector, and the answer generation model is used to generate the answer information matched by any question vector according to any question vector; According to the first question information or the first question vector, searching in the question answering database; The sorting model is invoked to sort the obtained pieces of answer information in descending order of matching degree with the first question information, and the answer information ranked first is determined as the target answer information. 2. The method according to claim 1, wherein the method further comprises: Obtain sample question information and corresponding sample answer information, as well as the sample matching degree between the sample question information and the sample answer information; According to the sample question information, the sample answer information and the sample matching degree, the ranking model is trained to obtain a trained ranking model. 3. The method according to claim 2, wherein the sorting model is trained according to the sample question information, the sample answer information and the sample matching degree to obtain a trained sorting model ,include: inputting the sample question information and the sample answer information into the ranking model, and obtaining the predicted matching degree between the sample question information and the sample answer information; According to the sample matching degree and the predicted matching degree, the ranking model is trained to obtain a trained ranking model. 4 . The method according to claim 1 , wherein the ranking model comprises a plurality of matching layers, a fusion layer and a ranking layer, and the calling ranking model is based on the degree of matching with the first question information. 5 . Sort the obtained pieces of answer information in order from high to low, and determine the first answer information as the target answer information, including: Calling the multiple matching layers to obtain the matching degree of each piece of answer information and the first question information respectively; Call the fusion layer to obtain the fusion matching degree of the multiple matching degrees of each piece of answer information; The sorting layer is called to sort the multiple pieces of answer information according to the order of fusion matching degree from high to low, and the answer information ranked first is determined as the target answer information. 5. The method according to claim 1, wherein the acquiring the first question vector of the first question information comprises: Perform word segmentation processing on the first question information to obtain a plurality of first words after word segmentation; obtaining similar words of the plurality of first words; Each time at least one first word is replaced with a corresponding similar word to generate a piece of similar question information; Obtain the first question vector according to the first question information and the at least one piece of similar question information. 6. The method according to claim 5, wherein the acquiring similar words of the plurality of first words comprises: Obtain, from a knowledge database, similar words associated with each of the plurality of first words, where the knowledge database is used to store similar words associated with each word; or, Obtain the similarity between each first word in the plurality of first words and at least one preset word, and use a preset word whose similarity with any first word is greater than the first preset similarity as the arbitrary first word. A similar word to the first word. 7. The method according to claim 5, wherein the obtaining the first question vector according to the first question information and the at least one piece of similar question information comprises: An average vector of the feature vector of the at least one piece of similar question information and the feature vector of the first question information is acquired as the first question vector. 8. The method according to claim 1, wherein the invoking the answer generation model, according to the first question vector and at least one second question vector, obtains at least one piece of answer information, comprising: splicing the first problem vector and the at least one second problem vector to obtain a third problem vector; The answer generation model is called, and the at least one piece of answer information is acquired according to the third question vector. 9. The method according to claim 1, wherein the retrieval in the question and answer database according to the first question information or the first question vector comprises: splicing the first problem vector and the at least one second problem vector to obtain a third problem vector; According to the similarity between the third question vector and the feature vector of each preset answer information in the question answering database, at least one piece of answer information is obtained, and the feature vector of the at least one piece of answer information is related to the third question The similarity of the vectors is greater than the similarity between the feature vectors of other preset answer information and the third question vector. 10. The method according to claim 1, wherein the at least one second question vector comprises a plurality of second question vectors, and the invoking answer generation model is based on the first question vector and the at least one second question vector. Question vector, get at least one answer information, including: performing a weighted average on the plurality of second problem vectors to obtain a fourth problem vector; Splicing the first problem vector and the fourth problem vector to obtain a fusion problem vector; According to the fusion question vector, the answer generation model is called to obtain the at least one piece of answer information. 11. The method according to claim 1, wherein the retrieval in the question and answer database according to the first question information or the first question vector comprises: performing a weighted average on the plurality of second problem vectors to obtain a fourth problem vector; Splicing the first problem vector and the fourth problem vector to obtain a fusion problem vector; At least one piece of answer information is obtained according to the similarity between the fusion question vector and the feature vector of each preset answer information in the question answering database, and the feature vector of the at least one piece of answer information is the difference between the fusion question vector and the fusion question vector. The similarity is greater than the similarity between the feature vector of other preset answer information and the fusion question vector. 12. The method according to claim 1, wherein the retrieval in the question and answer database according to the first question information or the first question vector comprises: According to the first question vector, perform retrieval in the question answering database to determine at least one fifth question vector, where the at least one fifth question vector includes question vectors similar to the first question vector; or, According to the first question vector and the at least one second question vector, perform retrieval in the question answering database to determine at least one fifth question vector, the at least one fifth question vector including the first question vector is similar At least one of the problem vectors of or the problem vectors that are similar to the at least one second problem vector; From the question and answer database, the answer information of the at least one fifth question vector is obtained as the at least one piece of answer information. 13. A question and answer device, characterized in that the device comprises: an acquisition module, configured to acquire the first question vector of the first question information and the second question vector of at least one piece of second question information, where the at least one piece of second question information is the question information acquired before the first question information; The generation module is used to call the answer generation model to obtain at least one piece of answer information according to the first question vector and at least one second question vector, and the answer generation model is used to generate the any question vector according to any question vector matching answer information; a retrieval module, configured to perform retrieval in the question and answer database according to the first question information or the first question vector; The sorting module is used to call the sorting model, sort the obtained multiple pieces of answer information in the order of matching degree with the first question information from high to low, and determine the answer information ranked first as the Describe the target answer information. 14. An electronic device, characterized in that the electronic device comprises one or more processors and one or more memories, wherein the one or more memories store at least one instruction, the at least one instruction is The one or more processors are loaded and executed to implement the operations performed by the question answering method of any one of claims 1 to 12. 15. A computer-readable storage medium, wherein at least one instruction is stored in the storage medium, and the at least one instruction is loaded and executed by a processor to implement the method according to any one of claims 1 to 12. The action performed by the Q&A method.

Images