WO2021159670A1

WO2021159670A1 - Method and apparatus for processing unknown question in intelligent questions and answers, computer device, and medium

Info

Publication number: WO2021159670A1
Application number: PCT/CN2020/105089
Authority: WO
Inventors: 范广
Original assignee: 深圳壹账通智能科技有限公司
Priority date: 2020-02-11
Filing date: 2020-07-28
Publication date: 2021-08-19
Also published as: CN111309881A

Abstract

Provided are a method and apparatus for processing an unknown question in intelligent questions and answers, a computer device, and a medium, which fall within the field of big data processing. The method comprises: receiving the current question input by a user and sent by a user terminal, and querying the current service tag corresponding to a stored previous question (S202); firstly, performing matching by means of a word library corresponding to the service tag; if the matching fails, continuing to perform matching by means of a slot word extraction model so as to determine whether the current question is an unknown question; when the current question is an unknown question, outputting the unknown question and the current service tag in an associated manner (S212); performing clustering processing on the output unknown question (S214); and sending the clustered unknown question to an operation terminal corresponding to the service tag (S216).

Description

Method, device, computer equipment and medium for processing unknown problems in intelligent question answering

Cross-references to related applications

This application claims the priority of a Chinese patent application filed with the Chinese Patent Office on February 11, 2020, the application number is 2020100872142, and the application name is "Methods, devices, computer equipment, and media for handling unknown questions in intelligent question and answering", and its entire contents Incorporated in this application by reference.

Technical field

This application relates to the field of artificial intelligence technology, and in particular to a method, device, computer equipment and medium for processing unknown questions in intelligent question answering.

Background technique

Frequently Asked Questions (FAQ) is a standard system for most products. It helps users find problems on their own to reduce the cost of manual customer service. The most common FAQ system on the market is a question-and-answer FAQ system. The user can directly in a chat dialog box. For consulting questions, the system provides an answer to similar questions based on keywords or the latest similarity algorithm.

Question-and-answer man-machine dialogue can provide users with a better user experience, but the inventor realizes that because a single round of similarity judgment will lead to problem matching is not necessarily very accurate (or because the accuracy of the system settings is high, it leads to finding Not a satisfactory answer). Therefore, user questions with unmatched answers will be generated in the background. If there are many users using the FAQ system, the set of unmatched questions will be very large, which will bring great pressure to the operation.

Summary of the invention

According to various embodiments disclosed in the present application, a method, device, computer equipment, and medium for processing unknown questions in intelligent question answering are provided.

A method for handling unknown questions in intelligent question answering, including:

Receive the current question entered by the user from the user terminal, and query the stored current service label corresponding to the previous question;

Acquiring the word database corresponding to the current business tag, performing word segmentation processing on the current question to obtain a number of word segmentation, matching the word segmentation with keywords in the thesaurus, and obtaining the number of successfully matched word segmentation;

When the number of successfully matched word segmentation is less than the preset value, the word segmentation is input into the pre-trained slot word extraction model to obtain the word corresponding to each word segmentation through the slot word extraction model type;

Acquiring the standard text corresponding to the word type, matching the standard text with keywords in the thesaurus, and determining the degree of relevance between the current question and the previous question according to the number of successfully matched standard texts;

When the degree of association is lower than a preset value, it is determined whether the current question is an unknown question according to preset rules; when the current question is an unknown question, the unknown question is associated with the current business tag Output

Clustering the output of the unknown problem; and

Send the clustered unknown question to the operation terminal corresponding to the service label.

A device for handling unknown questions in intelligent question answering, including:

The first receiving module is configured to receive the current question input by the user sent by the user terminal, and query the stored current service label corresponding to the previous question;

The word segmentation module is used to obtain the word database corresponding to the current business tag, perform word segmentation processing on the current question to obtain a number of word segmentation, match the word segmentation with the keywords in the thesaurus, and obtain the successfully matched word segmentation The number of

The slot word extraction module is used to input the word segmentation into a pre-trained slot word extraction model when the number of successfully matched word segmentation is less than a preset value, so as to be obtained through the slot word extraction model The word type corresponding to each said word segmentation;

The first relevance calculation module is used to obtain the standard text corresponding to the word type, match the standard text with the keywords in the thesaurus, and determine the current standard text according to the number of successfully matched standard texts. The degree of relevance of the question to the previous question;

The correlation output module is used to determine whether the current question is an unknown question according to preset rules when the degree of correlation is lower than a preset value; when the current question is an unknown question, combine the unknown question with Associated output of the current service label;

A clustering module for clustering the output of the unknown problem; and

The sending module is used to send the clustered unknown problem to the operation terminal corresponding to the service label.

A computer device, including a memory and one or more processors, the memory stores computer readable instructions, and when the computer readable instructions are executed by the processor, the one or more processors execute The following steps: receiving the current question entered by the user from the user terminal, and querying the stored current service label corresponding to the previous question;

Clustering the output of the unknown problem; and

One or more computer-readable storage media storing computer-readable instructions. When the computer-readable instructions are executed by one or more processors, the one or more processors perform the following steps:

Clustering the output of the unknown problem; and

The method, device, computer equipment, and medium for handling unknown questions in the above-mentioned intelligent question answering process, after receiving the current question entered by the user from the user terminal, first calculate the correlation between the current question and the previous question according to the current business label of the previous question, If the degree of relevance is lower than the threshold, it will converge through the convergence question library. If the question is still an unknown question after convergence, then the unknown question will be correlated with the current service label, that is, the current question that does not match the answer will be output. , That is, unknown issues are classified and sorted under their respective business labeling systems, turning a large number of single issues into a small number of multi-category issues. The operation terminal only needs to re-categorize the classification results or delete invalid issues. Yes, thereby improving the processing efficiency.

The details of one or more embodiments of the present application are set forth in the following drawings and description. Other features and advantages of this application will become apparent from the description, drawings and claims.

Description of the drawings

In order to more clearly describe the technical solutions in the embodiments of the present application, the following will briefly introduce the drawings needed in the embodiments. Obviously, the drawings in the following description are only some embodiments of the present application. A person of ordinary skill in the art can obtain other drawings based on these drawings without creative work.

Fig. 1 is an application scenario diagram of a method for processing unknown questions in intelligent question answering according to one or more embodiments.

Fig. 2 is a schematic flowchart of a method for processing unknown questions in intelligent question answering according to one or more embodiments.

Fig. 3 is a flowchart of the steps of calculating the degree of association according to one or more embodiments.

Fig. 4 is a structural block diagram of an unknown question processing device in intelligent question answering according to one or more embodiments.

Figure 5 is a block diagram of a computer device according to one or more embodiments.

Detailed ways

In order to make the technical solutions and advantages of the present application clearer, the following further describes the present application in detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described here are only used to explain the present application, and are not used to limit the present application.

The method for handling unknown questions in the intelligent question answering provided by this application can be applied to the application environment as shown in FIG. 1. Wherein, both the user terminal 102 and the operation terminal 106 can communicate with the server 104 through the network. The user terminal 102 can receive the current question entered by the user and send the current question to the server 104, so that the server 104 can query the stored current service label corresponding to the previous question, so that the server 104 obtains the word corresponding to the current service label Database, the current problem is segmented to obtain a number of word segmentation, then the word segmentation is matched with the keywords in the thesaurus, and the number of successfully matched word segments is obtained. When the number of successfully matched word segments is less than the preset value, Then input the word segmentation into the pre-trained slot word extraction model to obtain the word type corresponding to each word segmentation through the slot word extraction model; obtain the standard text corresponding to the word type, through the standard text and the keywords in the thesaurus Perform matching, and determine the degree of relevance between the current question and the previous question according to the number of successfully matched standard texts. When the calculated degree of relevance is lower than the preset value, the server 104 determines whether the current question is an unknown question according to the preset rule. If the server determines that the current problem is an unknown problem, it associates the unknown problem with the current service tag and outputs it. The server 104 clusters the unknown problem according to the service tag, and sends the clustered unknown problem to the operation terminal. In this way, unknown issues are classified and sorted under their respective business labeling systems, and a large number of single issues are turned into a small number of multi-category issues. The operation only needs to re-categorize the classification results or delete invalid issues to ensure the classification. Correctness. The user terminal 102 and the operation terminal 106 can be, but are not limited to, various personal computers, notebook computers, smart phones, tablet computers, and portable wearable devices. The server 104 can be implemented by an independent server or a server cluster composed of multiple servers. .

In one of the embodiments, as shown in FIG. 2, a method for processing unknown questions in intelligent question answering is provided. Taking the method applied to the server in FIG. 1 as an example for description, the method includes the following steps:

S202: Receive the current question input by the user sent by the user terminal, and query the stored current service label corresponding to the previous question.

Specifically, the user terminal first receives the question, that is, after the session is created, it sends the question to the server. The server determines whether the user terminal is creating a session. If so, the server creates this session accordingly, and then creates each common session. When a short answer to a question is asked, it is necessary to put a business label on the short answer to the common question to mark the business question of which business the short answer to the common question is for.

In this way, when the user inputs a question through the terminal, the server saves the current service label corresponding to the short answer to the most recent common question in real time.

Therefore, when the terminal receives a question newly input by the user, the server can obtain the current service label saved last time, that is, the service label of the short answer to the last common question saved last time.

S204: Obtain the thesaurus corresponding to the current business label, perform word segmentation processing on the current question to obtain a number of word segmentation, match the word segmentation with the keywords in the thesaurus, and obtain the number of successfully matched word segmentation.

When the server receives the current question, it determines the business field corresponding to the question according to the current business label, and then calculates the correlation between the current question and the question that has been asked in the business field. That is to say, the server only needs to calculate the current question and the question. The degree of relevance of the questions that have been asked in the business field reduces the scope of matching, and then performs matching, which can improve the efficiency of matching. There are two ways to calculate the correlation degree. One is the thesaurus matching method, that is, each business tag corresponds to a thesaurus, the server calculates the correlation between the current question and the words in the thesaurus, and the second method is training The method of the model, that is, when the word database is not successfully matched, that is, when the relevance is 0 or lower than the threshold, the server inputs the current question into the training model to obtain the slot word, and then uses the slot word corresponding to the current business label The lexicon is matched to get the degree of relevance.

Specifically, the keyword lexicon is a lexicon created and managed by the operators according to their own business characteristics. Therefore, the lexicon corresponds to a business tag, and a business tag corresponds to a lexicon, and the lexicon is pre-configured in the server background. The server can select the corresponding vocabulary according to the current business tag.

Specifically, the word segmentation process can use a traditional word segmentation algorithm, such as the NLP algorithm. The number of successfully matched tokens may refer to the number of successfully matched tokens under fuzzy matching.

S206: When the number of successfully matched word segmentation is less than the preset value, input the word segmentation into the pre-trained slot word extraction model, so as to obtain the word type corresponding to each word segmentation through the slot word extraction model.

Specifically, the slot extraction scheme requires pre-training of the model, that is, the association between words and word types obtained through training. The word types can include insurance types, insurance companies, and intended products. The server segmentation of the current question, and then input the segmentation into the model to obtain the word type corresponding to each segmentation.

S208: Obtain the standard text corresponding to the word type, match the standard text with the keywords in the lexicon, and determine the degree of relevance between the current question and the previous question according to the number of successfully matched standard texts.

Specifically, the standard text corresponding to the word type is obtained, and the standard text is matched with the keywords in the thesaurus, and the relevance degree is obtained according to the number of successfully matched keywords. For example, the number of successful matches is higher than that of the segmentation The total number can get the degree of relevance.

In practical applications, for example, the following dialogue:

User: Do you have this product?

System: Hello, we have this product. //Assuming that Children's Worry is a specific product name of the company, as a keyword has been configured in the thesaurus, and the keyword is mentioned in the user's question, the system directly records the keyword. And calculate the corresponding degree of relevance.

Slot extraction scheme (the second question is the same as this one). If the keyword vocabulary is not hit in the user's question, then the slot extraction algorithm is used to identify the user's intention, and after the slot is extracted, it will be similar to the user's question. Degree judgment.

Example:

User: I have Ping An Life Insurance, can I do a mortgage loan?

System: Personal mortgage loan qualification requirements are as follows: xxxxxxxx. //Assuming that no keywords are configured, the slot extraction algorithm can extract the following information "Insurance Type: Life Insurance", "Insurance Company: Ping An", "Intentional Product: Mortgage Loan" and then calculate the corresponding according to the extracted keywords The degree of relevance. In the foregoing embodiment, the degree of relevance is calculated in a variety of ways, which can improve the accuracy of calculating the degree of relevance. S210: When the degree of association is lower than the preset value, judge whether the current problem is an unknown problem according to the preset rule.

Specifically, when the degree of association is lower than the preset value, the server may continue to ask the user according to the preset rule to converge the current question, thereby judging whether the current question is an unknown question. For example, the current problem can be converged through the convergence problem library.

S212: When the current question is an unknown question, the unknown question is associated with the current service label and output. Specifically, if the server still does not confirm the business area of the current problem after converging the current problem according to the convergence problem database, that is, the problem is not stored in the corresponding business area, the server can mark the problem as an unknown problem and assign the unknown problem. The problem is output in association with the business tag, so that unknown problems under the specific business tag can be obtained. In this way, the unknown problems are associated with the current business tags, and the unknown problems are clustered, so that the operating terminal can see one type of one type of problem, rather than all the chaotic problems.

S214: Perform clustering processing on the output unknown problem.

Specifically, the clustering method here can adopt a traditional clustering method, for example, using similarity matching, that is, calculating the similarity of all questions, and taking the questions with high similarity as one category. The server clusters the unknown problems according to the current business tags, and divides the unknown problems with the same current business tags into one category.

S216: Send the clustered unknown question to the operation terminal corresponding to the service label.

Specifically, after clustering, different unknown problems in a business area are clustered. For example, in the financial management field, different types of problems such as repayment and purchase can be clustered, and the server will classify these different types of problems. The problem is returned to the operation terminal, so that the operation terminal can deal with problems of a certain category, instead of uniformly processing all unknown problems, which can improve the accuracy of processing.

Optionally, the server may calculate the number of unknown problems after clustering, and only send it to the operation after the preset number is reached, which facilitates the operation of one-time processing.

In the above-mentioned method for handling unknown questions in the smart question and answer, after receiving the current question entered by the user from the user terminal, first calculate the degree of relevance between the current question and the previous question according to the current service label of the previous question. If the degree of relevance is lower than the threshold, Then the unknown question is associated with the current business label and output, that is to say, the current question that does not match the answer, that is, the unknown question is sorted and sorted under the respective business label system, and a large number of single questions are turned into For a small number of multi-category issues, the operating terminal only needs to re-categorize or delete invalid issues based on the classification results, thereby improving the processing efficiency.

In one of the embodiments, the above method further includes: when the number of successfully matched word segmentation is greater than or equal to a preset value, determining the degree of relevance between the current question and the question that has been asked according to the data of the successfully matched word segmentation.

Specifically, the word segmentation is matched with the keywords in the thesaurus, and the degree of relevance between the current question and the question that has been asked is determined according to the data of the successfully matched word segmentation.

Specifically, the server matches the obtained word segmentation with the keywords in the thesaurus, and obtains the relevance degree according to the number of successfully matched keywords. For example, the relevance degree can be obtained by comparing the number of successfully matched words to the total number of word segmentation. .

In the foregoing embodiment, the degree of relevance is calculated in a variety of ways, which can improve the accuracy of calculating the degree of relevance.

In one of the embodiments, determining whether the current problem is an unknown problem according to preset rules includes:

S302: Extract the current convergence question from the convergence question database corresponding to the current service label, and send the current convergence question to the user terminal for display.

Specifically, the convergence question library corresponds to the service label, and the stored convergence question sentence is used to converge the question asked by the user to a certain type of question in a certain service area. There can be multiple convergence questions in the convergence question library. The server can first extract the current keywords from the current question, determine the meaning of the current keywords, and then extract the convergence questions from the convergence question database according to the meaning of the current keywords, and send the convergence questions to the user terminal for display, so that The user discovers the convergent question in time and answers it. The convergence question can be "Hello, confirm, you are still consulting [%s] (variables, fill in keywords or intentions)".

There are three situations when judging based on the degree of relevance:

The first is: if the keywords or intentions recorded in the previous round and the keywords or intentions extracted from the next round of user questions belong to two businesses, the keywords or intentions recorded in the previous round are directly discarded without convergence. For example, the last round is still about insurance, and the next round will directly ask the question about the credit card. Then, just use the next round of questions and go to the FAQ database for similarity analysis.

The second is: the intention of the next round of questions is not clear, you can randomly guide users to provide more information with unclearly directed questions, such as "Hello, I did not hear clearly, can you say something more", "Hello, I did not understand your question, can I trouble you to make it clearer?"

The third is: if there is a discrepancy between the intention of the next round of question and the intention of the previous round (the situation in this embodiment), the user is guided to clarify the question by bringing the vocabulary or slot into the question sentence. For example, "Hello, confirm, you are still consulting [%s] (variables, fill in keywords or intentions)".

S304: Receive a confirmation reply corresponding to the current convergence question returned by the user terminal, and match the business question corresponding to the service label according to the confirmation reply.

Specifically, after seeing the convergent question returned by the server, the user can confirm the convergent question or deny the convergent question. When the user gives a confirmation reply, the server can confirm the convergent question according to the keywords in the confirmation reply or the convergent question. The keywords in the sentences are matched with the business questions of the business tags.

S306: When the business question corresponding to the business label is not matched, mark the current question as an unknown question.

Specifically, if the server does not match the corresponding business problem, that is, the problem is not stored in the corresponding business area, the server can mark the problem as an unknown problem, and associate the unknown problem with the business label to output, so that it can be obtained Go to the unknown problem under the specific business tab. In this way, the unknown problems are associated with the current business tags, and the unknown problems are clustered, so that the operating terminal can see one type of one type of problem, rather than all the chaotic problems.

In the foregoing embodiment, the current problem is converged through the convergence problem library to determine whether the current problem is an unknown problem, and the processing efficiency is improved.

In one of the embodiments, extracting the current convergence question sentence from the convergence question database corresponding to the current business label includes: segmenting the current question to obtain representative words and the appearance order of the representative words; from the convergence question database corresponding to the current business label Select the initial question sentence corresponding to the representative word; select the initial question sentence whose vocabulary appearance order is consistent with the appearance order of the representative word as the current convergent question sentence.

Specifically, the server can perform word segmentation processing on the current problem, in which the word segmentation processing method can be the same as the above, and the sequence of the representative words obtained by the word segmentation can be obtained, so that the server can select and The initial question corresponding to the representative word. For example, the representative word first appears as the company, then the product type and finally the intended product, the server can match in this order to obtain the initial question, so that matching in order can improve the efficiency of matching .

For example, users: Do you have this product?

System: Hello, we have this product. //Assuming that Worry-free for Children is a specific product name of the company, as a keyword, it has been configured in the thesaurus. If the keyword is mentioned in the user's question, the system will directly record the keyword.

User: Does it protect children under 3 years old?

//At this time, if you directly use the question to do similarity in the answer database, the answer is basically not found or the probability of finding the answer is very low, because the intention is not clear. So you can use the key words recorded in the previous round to find the answer, and the similarity of the question is much higher.

In the slot extraction scheme, if the keyword vocabulary is not hit in the user's question, the slot extraction algorithm is used to identify the user's intention, and after the slot is extracted, the similarity judgment is made with the user's question.

Example:

User: I have Ping An Life Insurance, can I do a mortgage loan?

System: Personal mortgage loan qualification requirements are as follows: xxxxxxxx. //Assuming no keywords are configured, the slot extraction algorithm can extract the following information "Insurance Company: Ping An", "Insurance Type: Life Insurance", "Intentional Product: Mortgage Loan".

User: Is my insurance the qualification you require? //At this time, add the slot information extracted in the previous round to the user's intention and go to the answer database to query the answers to related questions.

In the above embodiment, the convergent question sentence is determined according to the appearance order of the keywords, which can improve the efficiency of determining the convergent question sentence.

In one of the embodiments, after sending the current convergent question to the user terminal for display, the method further includes: receiving the denial reply corresponding to the current convergent question returned by the user terminal, and judging whether the secondary denial reply can be extracted from the denial reply. Representative words; when the secondary representative words cannot be extracted from the denial reply, extract the next convergence question from the convergence question database as the current convergence question, and continue to send the current convergence question to the user terminal for display; until it is sent When the number of convergent questions to the user terminal reaches the preset value, or the user terminal does not receive a response corresponding to the current convergent question within the preset time period, it is determined that the current question is an unknown question, and the location problem is compared with the current business The label is associated with the output.

Specifically, receiving the denial reply corresponding to the current convergent question returned by the user terminal, and judging whether the secondary representative word can be extracted from the denial reply, and when the secondary representative word can be extracted from the denial reply, the second The high-level keywords are used as the current keywords, and the convergence questions are obtained from the convergence question library according to the current keywords, and then the convergence questions are sent to the terminal for display until the number of convergence questions reaches the preset value, or When the user stops the question and answer, the current question is output as an unknown question, and the unknown question is output in association with the current business label.

In the above-mentioned embodiment, the unknown questions are classified according to the replies of the users, so as to facilitate the subsequent processing of the terminal operation.

In one of the embodiments, after the clustered unknown question is sent to the operating terminal corresponding to the service tag, it includes: receiving the standard answer corresponding to the unknown question returned by the operating terminal, and corresponding the standard answer, the unknown question, and the unknown question After receiving the current question entered by the user from the user terminal and querying the current service label corresponding to the stored previous question, it also includes: according to the current service label, the current question is compared with the stored unknown question Matching; if the matching is successful, the standard response corresponding to the unknown question is obtained for output; otherwise, the correlation between the current question and the question that has been asked is continued to be calculated according to the current business label.

Specifically, after processing the unknown questions after clustering, the operating terminal can also add the standard answers, unknown questions, and business tags corresponding to the unknown questions to the corresponding library, that is, import the classified questions into the existing question library in batches. The processing efficiency can be greatly improved. That is, after receiving the current question, it is matched with the question library corresponding to the business label first, and if the matching is successful, the answer is directly output, otherwise the correlation between the current question and the question that has been asked is calculated according to the current business label.

In the above embodiment, after the operation has processed the unknown problem, a problem database of the unknown problem is established. The problem database is still classified according to the business tag, so that after receiving the current problem, the problem corresponding to the business tag can be directly queried. The answers in the library improve efficiency.

It should be understood that, although the various steps in the flowchart of FIGS. 2-3 are displayed in sequence as indicated by the arrows, these steps are not necessarily performed in sequence in the order indicated by the arrows. Unless specifically stated in this article, the execution of these steps is not strictly limited in order, and these steps can be executed in other orders. Moreover, at least some of the steps in Figure 2-3 may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily executed at the same time, but can be executed at different times. These sub-steps or stages The execution order of is not necessarily performed sequentially, but may be performed alternately or alternately with at least a part of other steps or sub-steps or stages of other steps.

In one of the embodiments, as shown in FIG. 4, a device for processing unknown questions in intelligent question answering is provided, including: a first receiving module 100, a word segmentation module 200, a slot word extraction module 300, The first correlation degree calculation module 400, the first judgment module 500, the correlation output module 600, the clustering module 700 and the sending module 800, wherein:

The first receiving module 100 is configured to receive the current question input by the user sent by the user terminal, and query the stored current service label corresponding to the previous question;

The word segmentation module 200 is used to obtain the vocabulary corresponding to the current business tag, perform word segmentation processing on the current question to obtain a number of word segmentation, match the word segmentation with keywords in the thesaurus, and obtain the number of successfully matched word segments;

The slot word extraction module 300 is used to input the word segmentation into the pre-trained slot word extraction model when the number of successfully matched word segmentation is less than the preset value, so as to obtain each word segmentation through the slot word extraction model Corresponding word type;

The first relevance calculation module 400 is used to obtain the standard text corresponding to the word type, match the standard text with the keywords in the lexicon, and determine the relevance of the current question to the previous question according to the number of successfully matched standard texts Spend;

The first judgment module 500 is used for judging whether the current problem is an unknown problem according to the preset rule when the degree of association is lower than the preset value;

The correlation output module 600 is used to correlate and output the unknown question with the current business label when the current question is an unknown question;

The clustering module 700 is used for clustering the output unknown problems;

The sending module 800 is configured to send the clustered unknown problem to the operation terminal corresponding to the service label.

In one of the embodiments, the device further includes:

The second degree of relevance calculation module is used to determine the degree of relevance between the current question and the question that has been asked according to the data of the successfully matched word segmentation when the number of successfully matched word segmentation is greater than or equal to the preset value.

In one of the embodiments, the correlation output module 600 may include:

The display unit is configured to extract the current convergence question sentence from the convergence question library corresponding to the current service label, and send the current convergence question sentence to the user terminal for display.

The receiving unit is configured to receive the confirmation reply corresponding to the current convergence question returned by the user terminal, and match the business question corresponding to the service label according to the confirmation reply.

The output unit is used to mark the current question as an unknown question when the service question corresponding to the service label is not matched.

In one of the embodiments, the above-mentioned display unit may include:

The sequence acquisition unit is used to segment the current question to obtain the representative words and the appearance order of the representative words.

The initial question selection unit is used to select the initial question corresponding to the representative word from the convergence question library corresponding to the current business label.

The convergent question selection unit is used to select the initial question sentence whose vocabulary appearance order is consistent with the appearance order of the representative word as the current convergent question sentence.

In one of the embodiments, the device for handling unknown questions in intelligent question answering may further include:

The third receiving module is used to receive the denial reply corresponding to the current convergent question returned by the user terminal, and determine whether the secondary representative words can be extracted from the denial reply.

The extraction module is used to extract the next convergence question from the convergence question database as the current convergence question when the secondary representative word cannot be extracted from the denial reply, and continue to send the current convergence question to the user terminal for display.

The loop module is used to determine that the current question is an unknown question until the number of convergent questions sent to the user terminal reaches a preset value, or the user terminal does not receive a response corresponding to the current convergent question within a preset time period, And correlate the location problem with the current business label and output it.

The fourth receiving module is used to receive the standard answer corresponding to the unknown question returned by the operation terminal, and store the standard answer, the unknown question, and the service label corresponding to the unknown question in association.

The matching module is used to match the current question with the stored unknown question according to the current business tag.

The output module is used to obtain the standard response corresponding to the unknown question for output if the matching is successful; otherwise, continue to calculate the correlation degree between the current question and the question that has been asked according to the current service label.

For the specific limitation of the unknown question processing device in the intelligent question and answer, please refer to the above limitation on the method of processing the unknown question in the intelligent question answering, which will not be repeated here. Each module in the device for handling unknown questions in the above intelligent question answering can be implemented in whole or in part by software, hardware, and a combination thereof. The above-mentioned modules may be embedded in the form of hardware or independent of the processor in the computer equipment, or may be stored in the memory of the computer equipment in the form of software, so that the processor can call and execute the operations corresponding to the above-mentioned modules.

In one of the embodiments, a computer device is provided. The computer device may be a server, and its internal structure diagram may be as shown in FIG. 5. The computer equipment includes a processor, a memory, a network interface, and a database connected through a system bus. Among them, the processor of the computer device is used to provide calculation and control capabilities. The memory of the computer device includes a non-volatile or volatile storage medium and internal memory. The non-volatile or volatile storage medium stores an operating system, computer readable instructions, and a database. The internal memory provides an environment for the operation of the operating system and computer-readable instructions in the non-volatile storage medium. The database of the computer equipment is used for data such as convergence problem database and business tags. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer readable instruction is executed by the processor, a method for processing unknown questions in intelligent question answering is realized.

Those skilled in the art can understand that the structure shown in FIG. 5 is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may Including more or fewer parts than shown in the figure, or combining some parts, or having a different arrangement of parts.

A computer device, including a memory and one or more processors. The memory stores computer-readable instructions. When the computer-readable instructions are executed by the processor, the one or more processors perform the following steps: The current question entered by the user, and query the stored current business label corresponding to the previous question; get the thesaurus corresponding to the current business label, perform word segmentation processing on the current question to obtain several word segmentation, and match the word segmentation with the keywords in the thesaurus , And obtain the number of successfully matched word segmentation; when the number of successfully matched word segmentation is less than the preset value, the word segmentation is input into the pre-trained slot word extraction model to obtain each word segmentation model through the slot word extraction model The word type corresponding to a word segmentation; obtain the standard text corresponding to the word type, match the standard text with the keywords in the thesaurus, and determine the degree of relevance between the current question and the previous question according to the number of successfully matched standard texts; When the degree of relevance is lower than the preset value, judge whether the current question is an unknown question according to the preset rules; when the current question is an unknown question, associate the unknown question with the current business label; and cluster the output unknown question Processing; and sending the clustered unknown problem to the operation terminal corresponding to the service label.

In one of the embodiments, the processor further implements the following steps when executing the computer-readable instructions: when the number of successfully matched word segmentation is greater than or equal to the preset value, the current question and the questioned question are determined according to the data of the successfully matched word segmentation The degree of relevance of the question.

In one of the embodiments, when the processor executes the computer-readable instructions, the process of judging whether the current question is an unknown question according to preset rules includes: extracting the current convergence question from the convergence question library corresponding to the current business tag, and adding The current convergence question is sent to the user terminal for display; the confirmation response corresponding to the current convergence question returned by the user terminal is received, and the service question corresponding to the service label is matched according to the confirmation response; and when the service corresponding to the service label is not matched When there is a question, the current question is marked as an unknown question.

In one of the embodiments, when the processor executes the computer-readable instruction, extracting the current convergence question from the convergence question library corresponding to the current service tag includes: segmenting the current question to obtain the representative words and the appearance order of the representative words ; Select the initial question sentence corresponding to the representative word from the convergence question library corresponding to the current business label; and select the initial question sentence whose vocabulary appearance order is consistent with the appearance order of the representative word as the current convergence question sentence.

In one of the embodiments, after the processor executes the computer-readable instruction to send the current convergent question to the user terminal for display, the method further includes: receiving a denial reply corresponding to the current convergent question returned by the user terminal, and Determine whether the secondary representative words can be extracted from the denial reply; when the secondary representative words cannot be extracted from the denial reply, extract the next convergence question from the convergence question database as the current convergence question, and continue to converge the current The question is sent to the user terminal for display; and until the number of convergent questions sent to the user terminal reaches a preset value, or the user terminal does not receive a reply corresponding to the current convergent question within a preset time period, it is determined that the current convergent question The problem is an unknown problem, and the location problem is associated with the current business label and output.

In one of the embodiments, after the processor executes the computer-readable instruction and sends the clustered unknown question to the operation terminal corresponding to the service tag, the method includes: receiving a standard response corresponding to the unknown question returned by the operation terminal , The standard answers, unknown questions, and business tags corresponding to the unknown questions are associated and stored; the processor executes the computer-readable instructions to receive the current question sent by the user from the user terminal and query the stored previous question corresponding to the current question After the current business label, it also includes: matching the current question with the stored unknown question according to the current business label; and if the matching is successful, obtain the standard answer corresponding to the unknown question and output it, otherwise, continue to calculate the current question based on the current business label The degree of relevance to the question that has been asked.

One or more computer-readable storage media storing computer-readable instructions. When the computer-readable instructions are executed by one or more processors, the one or more processors perform the following steps: receiving user input from a user terminal Current question, and query the stored current business tag corresponding to the previous question; get the thesaurus corresponding to the current business tag, perform word segmentation processing on the current question to obtain several word segmentation, match the word segmentation with the keywords in the thesaurus, and get The number of successfully matched tokens; when the number of successfully matched tokens is less than the preset value, the tokens are input into the pre-trained slot word extraction model to obtain the corresponding word segmentation through the slot word extraction model Type of words; get the standard text corresponding to the word type, match the standard text with the keywords in the lexicon, and determine the degree of relevance between the current question and the previous question according to the number of successfully matched standard texts; when the degree of relevance is low At the preset value, judge whether the current question is an unknown question according to the preset rules; when the current question is an unknown question, associate the unknown question with the current business label and output; cluster the output unknown question; and The clustered unknown questions are sent to the operation terminal corresponding to the service label.

Wherein, the computer-readable storage medium may be non-volatile or volatile.

In one of the embodiments, when the computer-readable instruction is executed by the processor, the following steps are also implemented: when the number of successfully matched word segmentation is greater than or equal to a preset value, the current problem and the existing problem are determined according to the data of the successfully matched word segmentation. The relevance of the question asked.

In one of the embodiments, when the computer-readable instruction is executed by the processor, judging whether the current question is an unknown question according to preset rules includes: extracting the current convergence question from the convergence question library corresponding to the current business tag, and Send the current convergence question to the user terminal for display; receive the confirmation reply corresponding to the current convergence question returned by the user terminal, and match the service question corresponding to the service label according to the confirmation response; and when the service label corresponding to the service label is not matched When there is a business problem, the current problem is marked as an unknown problem.

In one of the embodiments, the extraction of the current convergence question from the convergence question library corresponding to the current service tag when the computer-readable instruction is executed by the processor includes: segmenting the current question to obtain the representative word and the appearance of the representative word Sequence; select the initial question sentence corresponding to the representative word from the convergence question library corresponding to the current business label; and select the initial question sentence whose vocabulary appearance order is consistent with the appearance order of the representative word as the current convergent question sentence.

In one of the embodiments, after the computer-readable instruction is executed by the processor to send the current convergent question to the user terminal for display, the method further includes: receiving a denial reply corresponding to the current convergent question returned by the user terminal, And judge whether the secondary representative words can be extracted from the denial reply; when the secondary representative words cannot be extracted from the denial reply, the next convergent question sentence is extracted from the convergent question database as the current convergent question sentence, and the current convergent question is continued. The convergent question is sent to the user terminal for display; and until the number of convergent questions sent to the user terminal reaches the preset value, or the user terminal does not receive a reply corresponding to the current convergent question within the preset time period, it is determined The current problem is an unknown problem, and the location problem is associated with the current business label and output.

In one of the embodiments, when the computer-readable instruction is executed by the processor, after the clustered unknown problem is sent to the operation terminal corresponding to the service tag, it includes: receiving the standard corresponding to the unknown problem returned by the operation terminal Reply: Associate and store standard answers, unknown questions, and business tags corresponding to unknown questions; when the computer-readable instructions are executed by the processor, it receives the current question entered by the user from the user terminal and queries the stored previous question After the corresponding current business label, it also includes: matching the current question with the stored unknown question according to the current business label; and if the matching is successful, obtain the standard answer corresponding to the unknown question for output, otherwise, continue to calculate based on the current business label The degree of relevance of the current question to the question that has been asked.

A person of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be implemented by instructing relevant hardware through computer-readable instructions. The computer-readable instructions can be stored in a computer-readable storage. In the medium, when the computer-readable instructions are executed, they may include the processes of the above-mentioned method embodiments. Wherein, any reference to memory, storage, database or other media used in the embodiments provided in this application may include non-volatile and/or volatile memory. Non-volatile memory may include read only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. As an illustration and not a limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous chain Channel (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

The technical features of the above embodiments can be combined arbitrarily. In order to make the description concise, all possible combinations of the technical features in the above embodiments are not described. However, as long as there is no contradiction in the combination of these technical features, they should all be combined. It is considered as the range described in this specification.

The above-mentioned embodiments only express several implementation manners of the present application, and the description is relatively specific and detailed, but it should not be understood as a limitation on the scope of the invention patent. It should be pointed out that for those of ordinary skill in the art, without departing from the concept of this application, several modifications and improvements can be made, and these all fall within the protection scope of this application. Therefore, the scope of protection of the patent of this application shall be subject to the appended claims.

Claims

A method for handling unknown questions in intelligent question answering, including:

Receive the current question entered by the user from the user terminal, and query the stored current service label corresponding to the previous question;

Acquiring the word database corresponding to the current business tag, performing word segmentation processing on the current question to obtain a number of word segmentation, matching the word segmentation with keywords in the thesaurus, and obtaining the number of successfully matched word segmentation;

When the number of successfully matched word segmentation is less than the preset value, the word segmentation is input into the pre-trained slot word extraction model to obtain the word corresponding to each word segmentation through the slot word extraction model type;

Acquiring the standard text corresponding to the word type, matching the standard text with keywords in the thesaurus, and determining the degree of relevance between the current question and the previous question according to the number of successfully matched standard texts;

When the degree of association is lower than a preset value, it is determined whether the current question is an unknown question according to preset rules; when the current question is an unknown question, the unknown question is associated with the current business tag Output

Clustering the output of the unknown problem; and

Send the clustered unknown question to the operation terminal corresponding to the service label.
The method according to claim 1, wherein the method further comprises:

When the number of successfully matched word segmentation is greater than or equal to the preset value, the degree of relevance between the current question and the question that has been asked is determined according to the data of the successfully matched word segmentation.
The method according to claim 2, wherein the judging whether the current problem is an unknown problem according to a preset rule comprises:

Extracting the current convergence question sentence from the convergence question database corresponding to the current service label, and sending the current convergence question sentence to the user terminal for display;

Receiving a confirmation reply corresponding to the current convergence question returned by the user terminal, and matching the business question corresponding to the service label according to the confirmation reply; and

When the service question corresponding to the service label is not matched, the current question is marked as an unknown question.
The method according to claim 3, wherein said extracting the current convergence question from the convergence question database corresponding to the current service label comprises:

Perform word segmentation on the current question to obtain representative words and the appearance order of the representative words;

Selecting the initial question sentence corresponding to the representative word from the convergence question library corresponding to the current business label; and

The initial question sentence whose vocabulary appearance order is consistent with the appearance order of the representative words is selected as the current convergent question sentence.
The method according to claim 4, wherein after the sending the current convergence question to the user terminal for display, the method further comprises:

Receiving a denial reply corresponding to the current convergence question returned by the user terminal, and determining whether secondary representative words can be extracted from the denial reply;

When the secondary representative word cannot be extracted from the denial reply, the next convergence question is extracted from the convergence question library as the current convergence question, and the current convergence question is continued to be sent to the user terminal Display; and

Until the number of convergent questions sent to the user terminal reaches a preset value, or the user terminal does not receive a response corresponding to the current convergent question within a preset time period, it is determined that the current question is Unknown problem, and correlate and output the location problem with the current service label.
The method according to any one of claims 1 to 5, wherein after the clustered unknown question is sent to the operating terminal corresponding to the service label, the method further comprises:

Receiving the standard answer corresponding to the unknown question returned by the operation terminal, and storing the standard answer, the unknown question, and the service label corresponding to the unknown question in association;

After receiving the current question input by the user sent by the user terminal and querying the stored current service label corresponding to the previous question, the method further includes:

Matching the current question with the stored unknown question according to the current service tag; and

If the matching is successful, obtain the standard response corresponding to the unknown question and output it; otherwise, continue to calculate the degree of relevance between the current question and the question that has been asked according to the current service tag.
A device for handling unknown questions in intelligent question answering, including:

The first receiving module is configured to receive the current question input by the user sent by the user terminal, and query the stored current service label corresponding to the previous question;

The word segmentation module is used to obtain the word database corresponding to the current business tag, perform word segmentation processing on the current question to obtain a number of word segmentation, match the word segmentation with the keywords in the thesaurus, and obtain the successfully matched word segmentation The number of

The slot word extraction module is used to input the word segmentation into a pre-trained slot word extraction model when the number of successfully matched word segmentation is less than a preset value, so as to be obtained through the slot word extraction model The word type corresponding to each said word segmentation;

The first relevance calculation module is used to obtain the standard text corresponding to the word type, match the standard text with the keywords in the thesaurus, and determine the current standard text according to the number of successfully matched standard texts. The degree of relevance of the question to the previous question;

The correlation output module is used to determine whether the current question is an unknown question according to preset rules when the degree of correlation is lower than a preset value; when the current question is an unknown question, combine the unknown question with Associated output of the current service label;

A clustering module for clustering the output of the unknown problem; and

The sending module is used to send the clustered unknown problem to the operation terminal corresponding to the service label.
The device according to claim 7, wherein the device further comprises:

The second relevance calculation module is configured to determine the relevance between the current question and the question that has been asked according to the data of the successfully matched word segmentation when the number of successfully matched word segmentation is greater than or equal to the preset value.
A computer device includes a memory and one or more processors. The memory stores computer-readable instructions. When the computer-readable instructions are executed by the one or more processors, the one or more Each processor performs the following steps:

Receive the current question entered by the user from the user terminal, and query the stored current service label corresponding to the previous question;

Acquiring the word database corresponding to the current business tag, performing word segmentation processing on the current question to obtain a number of word segmentation, matching the word segmentation with keywords in the thesaurus, and obtaining the number of successfully matched word segmentation;

When the number of successfully matched word segmentation is less than the preset value, the word segmentation is input into the pre-trained slot word extraction model to obtain the word corresponding to each word segmentation through the slot word extraction model type;

Acquiring the standard text corresponding to the word type, matching the standard text with keywords in the thesaurus, and determining the degree of relevance between the current question and the previous question according to the number of successfully matched standard texts;

When the degree of association is lower than a preset value, it is determined whether the current question is an unknown question according to preset rules; when the current question is an unknown question, the unknown question is associated with the current business tag Output

Clustering the output of the unknown problem; and

Send the clustered unknown question to the operation terminal corresponding to the service label.
The computer device according to claim 9, wherein the processor further executes the following steps when executing the computer readable instruction:

When the number of successfully matched word segmentation is greater than or equal to the preset value, the degree of relevance between the current question and the question that has been asked is determined according to the data of the successfully matched word segmentation.
11. The computer device according to claim 10, wherein the determining whether the current problem is an unknown problem according to a preset rule, which is implemented when the processor executes the computer-readable instruction, comprises:

Extracting the current convergence question sentence from the convergence question database corresponding to the current service label, and sending the current convergence question sentence to the user terminal for display;

Receiving a confirmation reply corresponding to the current convergence question returned by the user terminal, and matching the business question corresponding to the service label according to the confirmation reply; and

When the service question corresponding to the service label is not matched, the current question is marked as an unknown question.
11. The computer device according to claim 11, wherein the extracting the current convergence question from the convergence question database corresponding to the current service tag, which is implemented when the processor executes the computer-readable instruction, comprises:

Perform word segmentation on the current question to obtain representative words and the appearance order of the representative words;

Selecting the initial question sentence corresponding to the representative word from the convergence question library corresponding to the current business label; and

The initial question sentence whose vocabulary appearance order is consistent with the appearance order of the representative words is selected as the current convergent question sentence.
The computer device according to claim 12, wherein, after the current convergence question is sent to the user terminal for display by the processor when the processor executes the computer-readable instruction, the method further include:

Receiving a denial reply corresponding to the current convergence question returned by the user terminal, and determining whether secondary representative words can be extracted from the denial reply;

When the secondary representative word cannot be extracted from the denial reply, the next convergence question is extracted from the convergence question library as the current convergence question, and the current convergence question is continued to be sent to the user terminal Display; and

Until the number of convergent questions sent to the user terminal reaches a preset value, or the user terminal does not receive a response corresponding to the current convergent question within a preset time period, it is determined that the current question is Unknown problem, and correlate and output the location problem with the current service label.
The computer device according to any one of claims 9 to 13, wherein the said processor executes the computer-readable instruction to send the clustered unknown question to the corresponding service tag After operating the terminal, it also includes:

Receiving the standard answer corresponding to the unknown question returned by the operation terminal, and storing the standard answer, the unknown question, and the service label corresponding to the unknown question in association;

When the processor executes the computer-readable instruction, after the receiving the current question input by the user sent by the user terminal and querying the stored current service tag corresponding to the previous question, the method further includes:

Matching the current question with the stored unknown question according to the current service tag; and

If the matching is successful, obtain the standard response corresponding to the unknown question and output it; otherwise, continue to calculate the degree of relevance between the current question and the question that has been asked according to the current service tag.
One or more computer-readable storage media storing computer-readable instructions, which when executed by one or more processors, cause the one or more processors to perform the following steps:

Receive the current question entered by the user from the user terminal, and query the stored current service label corresponding to the previous question;

Acquiring the word database corresponding to the current business tag, performing word segmentation processing on the current question to obtain a number of word segmentation, matching the word segmentation with keywords in the thesaurus, and obtaining the number of successfully matched word segmentation;

When the number of successfully matched word segmentation is less than the preset value, the word segmentation is input into the pre-trained slot word extraction model to obtain the word corresponding to each word segmentation through the slot word extraction model type;

Acquiring the standard text corresponding to the word type, matching the standard text with keywords in the thesaurus, and determining the degree of relevance between the current question and the previous question according to the number of successfully matched standard texts;

When the degree of association is lower than a preset value, it is determined whether the current question is an unknown question according to preset rules; when the current question is an unknown question, the unknown question is associated with the current business tag Output

Clustering the output of the unknown problem; and

Send the clustered unknown question to the operation terminal corresponding to the service label.
The storage medium according to claim 15, wherein the following steps are further performed when the computer-readable instructions are executed by the processor:

When the number of successfully matched word segmentation is greater than or equal to the preset value, the degree of relevance between the current question and the question that has been asked is determined according to the data of the successfully matched word segmentation.
16. The storage medium according to claim 16, wherein the determining whether the current problem is an unknown problem according to a preset rule, which is implemented when the computer-readable instruction is executed by the processor, comprises:

Extracting the current convergence question sentence from the convergence question database corresponding to the current service label, and sending the current convergence question sentence to the user terminal for display;

Receiving a confirmation reply corresponding to the current convergence question returned by the user terminal, and matching the business question corresponding to the service label according to the confirmation reply; and

When the service question corresponding to the service label is not matched, the current question is marked as an unknown question.
18. The storage medium according to claim 17, wherein the extracting the current convergence question from the convergence question database corresponding to the current service tag, which is implemented when the computer-readable instruction is executed by the processor, comprises:

Perform word segmentation on the current question to obtain representative words and the appearance order of the representative words;

Selecting the initial question sentence corresponding to the representative word from the convergence question library corresponding to the current business label; and

The initial question sentence whose vocabulary appearance order is consistent with the appearance order of the representative words is selected as the current convergent question sentence.
18. The storage medium according to claim 18, wherein after the sending of the current convergence question to the user terminal for display by the computer-readable instruction when the computer-readable instruction is executed by the processor, the method Also includes:

Receiving a denial reply corresponding to the current convergence question returned by the user terminal, and determining whether secondary representative words can be extracted from the denial reply;

When the secondary representative word cannot be extracted from the denial reply, extract the next convergence question from the convergence question library as the current convergence question, and continue to send the current convergence question to the user terminal Display; and

Until the number of convergent questions sent to the user terminal reaches a preset value, or the user terminal does not receive a response corresponding to the current convergent question within a preset time period, it is determined that the current question is Unknown problem, and correlate and output the location problem with the current service label.
The storage medium according to any one of claims 15 to 19, wherein when the computer-readable instruction is executed by the processor, the clustered unknown problem is sent to the corresponding service label After operating the terminal, it also includes:

Receiving the standard answer corresponding to the unknown question returned by the operation terminal, and storing the standard answer, the unknown question, and the service label corresponding to the unknown question in association;

When the computer-readable instruction is executed by the processor, after the receiving the current question entered by the user from the user terminal and querying the stored current service tag corresponding to the previous question, the method further includes:

Matching the current question with the stored unknown question according to the current service tag; and

If the matching is successful, obtain the standard response corresponding to the unknown question and output it; otherwise, continue to calculate the degree of relevance between the current question and the question that has been asked according to the current service tag.