CN113111157A

CN113111157A - Question-answer processing method, device, computer equipment and storage medium

Info

Publication number: CN113111157A
Application number: CN202110354868.1A
Authority: CN
Inventors: 刘玉; 李松如; 文博; 刘云峰
Original assignee: Shenzhen Zhuiyi Technology Co Ltd
Current assignee: Shenzhen Zhuiyi Technology Co Ltd
Priority date: 2021-03-31
Filing date: 2021-03-31
Publication date: 2021-07-13
Anticipated expiration: 2041-03-31
Also published as: CN113111157B

Abstract

The application relates to a question and answer processing method, a question and answer processing device, computer equipment and a storage medium. The method comprises the following steps: acquiring a historical conversation corresponding to an unresolved user question; determining a corresponding task flow according to the intention of the historical conversation; determining a return visit clarifying operation corresponding to the unresolved user question according to the task flow; acquiring a user access channel corresponding to the unresolved user question; and returning the user who provides the unresolved user question on the basis of the return visit clarification word through the user access channel so as to complete the task flow. By adopting the method, the service pressure of manual customer service can be relieved.

Description

Question-answer processing method, device, computer equipment and storage medium

Technical Field

The present application relates to the field of artificial intelligence technologies, and in particular, to a question and answer processing method and apparatus, a computer device, and a storage medium.

Background

And the intelligent customer service guides the user to provide word slot and intention information through clarification technology according to the task flow chart and the state of word slot and intention collection in the conversation, and the whole task processing flow is completed. In the conversation process, the intention is fuzzy due to unclear description of the user, or the recognition accuracy of a certain intention or word slot of the intelligent customer service on certain nodes of certain tasks is low, so that the intelligent customer service gives no questions, unsolved problems are accumulated in a short period, and the manual customer service pressure is high.

Disclosure of Invention

In view of the above, it is necessary to provide a question and answer processing method, apparatus, computer device and storage medium for the above technical problems.

A question-answer processing method, the method comprising:

acquiring a historical conversation corresponding to an unresolved user question;

determining a corresponding task flow according to the intention of the historical conversation;

determining a return visit clarifying operation corresponding to the unresolved user question according to the task flow;

acquiring a user access channel corresponding to the unresolved user question;

and returning the user who provides the unresolved user question on the basis of the return visit clarification word through the user access channel so as to complete the task flow. A question-answering processing apparatus, the apparatus comprising:

the historical conversation acquisition module is used for acquiring the historical conversation corresponding to the unresolved user question;

the task flow determining module is used for determining a corresponding task flow according to the intention of the historical conversation;

a clarification word operation determining module, configured to determine a return visit clarification word operation corresponding to the unresolved user question according to the task flow;

the return visit module is used for acquiring a user access channel corresponding to the unresolved user question;

the return visit module is also used for returning visits to the users who propose the unresolved user question based on the return visit clarification technique through the user access channel so as to complete the task flow.

A computer device comprising a memory and a processor, the memory storing a computer program, the processor implementing the following steps when executing the computer program: acquiring a historical conversation corresponding to an unresolved user question;

acquiring a user access channel corresponding to the unresolved user question;

and returning the user who provides the unresolved user question on the basis of the return visit clarification word through the user access channel so as to complete the task flow.

A computer-readable storage medium, on which a computer program is stored which, when executed by a processor, carries out the steps of:

acquiring a user access channel corresponding to the unresolved user question;

According to the question-answer processing method, the question-answer processing device, the computer equipment and the storage medium, more accurate return visit clarification can be obtained by acquiring the historical conversation corresponding to the unresolved user question, determining the corresponding task flow according to the intention of the historical conversation, and determining the return visit clarification according to the task flow; the method comprises the steps of obtaining a user access channel corresponding to an unresolved user question, and actively returning to the unresolved robot through the user access channel, so that the confidence of a user on intelligent customer service capacity can be improved, the problem of the user can be actively solved, customer service pressure is relieved, and user experience is improved.

Drawings

FIG. 1 is a diagram of an application environment of a question and answer processing method in one embodiment;

FIG. 2 is a schematic flow chart diagram illustrating a method for question and answer processing in one embodiment;

FIG. 3 is a schematic flow chart diagram of a question-answer processing method in another embodiment;

FIG. 4 is a block diagram showing the structure of a question answering processing apparatus in one embodiment;

FIG. 5 is a diagram illustrating an internal structure of a computer device according to an embodiment.

Detailed Description

In order to make the objects, technical solutions and advantages of the present application more apparent, the present application is described in further detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the present application and are not intended to limit the present application.

The question answering processing method provided by the application can be applied to the application environment shown in fig. 1. Wherein the terminal 102 communicates with the server 104 via a network. The terminal 102 may be, but not limited to, various personal computers, notebook computers, smart phones, tablet computers, and portable wearable devices, and the server 104 may be implemented by an independent server or a server cluster formed by a plurality of servers. The server 104 acquires a historical conversation corresponding to the unresolved user question; determining a corresponding task flow according to the intention of the historical conversation; determining a return visit clarifying operation corresponding to the unresolved user question according to the task flow; acquiring a user access channel corresponding to an unresolved user question; and returning the users who propose unresolved user question sentences based on a return visit clarification technique through the user access channel to complete the task flow. The terminal 102 is used for receiving the return visit clarifying dialog sent by the server. It is to be understood that the question answering method in the embodiment of the present application may be applied only to the server 104, or may be applied to an application environment formed by the terminal 102 and the server 104 shown in fig. 1.

In one embodiment, as shown in fig. 2, a question-answering processing method is provided, which is described by taking the example that the method is applied to the server in fig. 1, and includes the following steps:

step 202, obtaining the historical conversation corresponding to the unresolved question of the user.

The unresolved user question may be a question for which the server cannot recognize an intention. The unresolved user question may be a question of the same knowledge point category or a question of a different knowledge point category. The number of unresolved user questions is not limited. The unresolved question of the user may be expressed in text or voice.

The history session corresponding to the unresolved user question is a history session in which the server communicates with the terminal when the unresolved user question is generated.

Specifically, when the service quality evaluation value of the intelligent customer service is lower than a preset evaluation threshold value, the server acquires unresolved user question sentences within a preset time period and acquires historical conversations corresponding to the unresolved user question sentences.

And step 204, determining a corresponding task flow according to the intention of the history conversation.

And task flows corresponding to all intentions are preset in the server. For example, intent 1 corresponds to task flow a and intent 2 corresponds to task flow B. The task process comprises at least one task process node. A task flow node may represent a robot question, i.e. a return visit clarification phrase.

Specifically, the server identifies the intention of the historical conversation based on the historical conversation, and determines the corresponding task flow based on the preset mapping relation between the intention and the task flow and the intention of the historical conversation.

And step 206, determining a return visit clarifying statement corresponding to the unresolved user question according to the task flow.

Among them, the revisit clarification technique is used for inquiry, clarification and confirmation to collect information necessary for transacting business to a user. The return visit clarification session may be a statement. Revisiting clarification is a message related to an unresolved user question.

Specifically, the server obtains a return visit clarifying statement corresponding to each task flow node in the task flow according to the task flow, and obtains a return visit clarifying statement corresponding to an unresolved user question.

And step 208, acquiring a user access channel corresponding to the unresolved user statement.

The user access channel refers to a channel for a user to access a session. For example, user access channels include, but are not limited to, instant messaging applications, web pages, telephones.

Specifically, the server obtains a user access channel corresponding to the unresolved user question.

And step 210, returning visits to the users who provide unresolved user questions based on a return visit clarification technique through the user access channel to complete the task flow.

The statement at the time of return visit may be, for example, a statement for asking the user whether the unresolved user question has been made or not, or a statement for informing the user that the knowledge point classification model has been updated is not limited thereto.

Through the user access channel, the server visits the user who provides the unresolved user question based on the revisit clarification word, sends a message related to the unresolved user question to the user, and guides the user to respond to the sent revisit clarification word so as to solve the problem in the unresolved user question. The user access channel is taken as an example for telephone access. And if the user accesses the intelligent customer service robot through the telephone, the intelligent customer service robot dials back the user telephone. The robot revisiting function changes passivity into initiative, a fixed telephone is generated according to a template, and a revisiting clarification telephone is actively sent out, such as 'you are good, i are the intelligent customer service robot small w of a certain company, I realize upgrading optimization, and a certain problem is consulted before you to ask I to serve your again'. Some of these questions is a user question.

In this embodiment, during the return visit of the unsolved problem, the robot performs advance judgment on the entry intention of the task according to the user history session in which the unsolved problem is generated. Such as where the robot determines that the intent of the historical session is "billing query". The robot generates a return visit clarification session through the template, such as a robot inquiry of "do you need to inquire bills? ". Obtaining positive or negative intention according to the answer of the user, such as according to 'yes, i want to ask' obtaining positive intention; the negative intent is obtained by "not, you have made a mistake". Assuming that the user answers positively, the robot enters a task flow of 'bill inquiry', and the clarification dialect set by the task flow is sent to the user.

According to the question-answer processing method in the embodiment, the historical conversation corresponding to the unresolved user question is obtained, the corresponding task flow is determined according to the intention of the historical conversation, the return visit clarification dialogue is determined according to the task flow, the return visit can be carried out based on the intention of the user, and the more accurate return visit clarification dialogue is obtained; the method comprises the steps of obtaining a user access channel corresponding to an unresolved user question, and actively returning to the unresolved robot through the user access channel, so that the confidence of a user on intelligent customer service capacity can be improved, the problem of the user can be actively solved, customer service pressure is relieved, and user experience is improved.

In one embodiment, the method further comprises: acquiring a user reply sentence obtained by revisiting a clarifying dialog through a user access channel; extracting word slots and intention information from the user reply sentences; performing information query according to the word slot and the intention information to obtain an information query result; and returning an information query result through the user access channel.

Wherein, in the dialog system, the intentions and word slots collectively represent the results of the dialog system's understanding of the user's dialog. Here we assume that some necessary constraints in the user dialog need to be understood by the system and affect the result of the execution of the dialog system. This constraint is called a slot (slot). For example: the "central station" in the "zapping" is a "tv channel word slot" which will to some extent influence the system's execution of the "zapping" intention. The word slot is composed of a word slot category and a value, and the word slot category is mainly used for helping a developer to classify and process the definition conditions. Optionally, the word slot is also a condition required by the user, for example, in a "scene of inquiring weather", the system needs to obtain two conditions of time and place to inquire weather, and then "time" and "place" are the word slot.

Specifically, a user reply statement replied by the user to the return visit clarification session is obtained through the user access channel. And the server extracts the word slot and the intention information from the user reply sentence in real time, and performs information query according to the word slot and the intention information to obtain an information query result. And continuing to execute the step of revisiting the user who provides the unresolved user question based on the revisiting clarification session until the user finishes the session or outputs the last revisiting clarification session, and finishing the task flow.

In this embodiment, for example, the return call clarification term "ask for you to inquire about the current bill, the historical bill, or not bill? After hearing the return visit clarification session, the user gives a user reply sentence of 'amount, i want to inquire the bill at this stage', the robot recognizes the word slot example of 'bill at this stage' from the user reply sentence, and fills the word slot of 'bill type' with the example. Through the multiple rounds of question answering, task intention information of ' bill inquiry ' and word slot information of ' bill at this time ' are obtained, information is inquired for a user through a preset service interface, and the information is fed back to the user through dialect, if ' you inquire, you owe 99 yuan in the bill in 8 months. ". To this end, the robot has transacted the business for the user.

In the question and answer processing method in this embodiment, the user reply sentences obtained by revisiting the clarifying dialect are obtained, the word slots and the intention information are extracted from the user reply sentences, information query is performed according to the word slots and the intention information, an information query result is obtained, and the information query result is returned through the user access channel, so that good communication with the user can be established, the user problem is actively helped to be solved, the problem is solved, and the user experience is improved.

In one embodiment, the question-answering processing method further includes: acquiring the maximum return visit times of the user on the condition of not replying; and when the number of times of visiting the user back through the user access channel reaches the maximum number of times of visiting back, marking the unresolved user problem, and suspending processing of the unresolved user question.

The user does not reply may be, but not limited to, the user hanging up the phone, the user not replying to the message within a preset time, or the like. The maximum number of revisits refers to the maximum number of times that a user is contacted through the user access channel.

Specifically, the server obtains the maximum number of revisits of the same unresolved user question for the same user access channel. When the number of times of visiting the user back through the user access channel reaches the maximum number of times of visiting back, the unresolved user question is marked as the user not replying, and the unresolved user question is processed in a suspended mode or is abandoned.

In the question and answer processing method in this embodiment, the maximum number of times of return visits to the user without reply is obtained, and when the number of times of return visits to the user through the user access channel reaches the maximum number of times of return visits, the unresolved user question is marked, and the unresolved user question is processed in a deferred manner, so that resource waste is reduced.

In one embodiment, the method for completing the task flow by revisiting the user who provides the unresolved user question based on the revisiting clarification technique through the user access channel comprises the following steps:

and returning the users who propose unresolved user question sentences based on a return visit clarification technique in the idle time period of the users through the user access channel to finish the task flow.

The user idle time period may be an idle time period preset by the server. The user idle period may be a period in which the user actively seeks the robot. The idle time period may be, for example, lunch time, dinner time, non-working day time, and the like.

Specifically, the server accesses the user who provides the unresolved user question based on the return access clarification technique in the user idle time period through the user access channel to complete the task flow.

In the question-answer processing method in this embodiment, the user who presents the unresolved user question is revisited based on the revisiting clarification technique to complete the task flow within the user idle time period through the user access channel, so that convenience and possibility of answering by the user can be improved to solve the unresolved user question.

In one embodiment, the question-answering processing method further includes: when the emotion of the user is recognized to be a preset poor emotion, corresponding words of relieving the emotion are obtained;

and after outputting the words of the relaxing emotion, performing return visit on the users who propose the unresolved user question based on the return visit clarification words through the user access channel to finish the task flow.

The preset bad emotion may be an angry emotion or a vexation emotion. Mood-relieving dialects are used to keep users happy.

Specifically, when the server identifies that the emotion of the user is a preset poor emotion through the emotion analysis model, corresponding words of the relieved emotion are acquired. And after outputting the words of the relaxing emotion, performing return visit on the users who propose the unresolved user question based on the return visit clarification words through the user access channel to finish the task flow. The output can be through the characters output, also can be through the pronunciation output.

For example, if the user is uncooperative or abusive, the robot determines the emotion of the user according to the result of the emotion analysis model, and if the emotion of the user is identified to be poor, an optional technique of relieving emotion is given, and then the previous clarification technique is repeated to enable the user to provide mission-critical information. Such as a robot asking "do you need to query for a bill? "the user answers" you are fool ". Then "the emotion corresponding to the simple emotion" belongs to the preset poor emotion, the robot can output the corresponding emotion-relieving word "in contradistinction, i tries to improve, please give i a little time", and continue to perform the return visit to the user who has provided the unresolved user question based on the return visit clarification word "do you need to query the bill? ".

In the question-answer processing method in this embodiment, when it is recognized that the emotion of the user is a preset poor emotion, a corresponding emotion-relieving jargon is acquired, and after the emotion-relieving jargon is output, return visit is performed on the user who presents an unresolved user question based on a return visit clarifying jargon through a user access channel to complete a task flow, so that the processing efficiency of the user question can be improved and the user experience can be improved under the conditions that the user is not matched or abused.

In one embodiment, the question-answering processing method further includes: when a user reply statement which is not matched with the current return visit clarification word is detected, replying based on the unmatched user reply statement; after replying based on the unmatched user reply statement, the current return visit clarifying statement is output.

In particular, a user reply statement that does not match the return visit clarification session may refer to a user reply statement that is not related to the return visit clarification session. When a user reply statement which is not matched with the current return visit clarification word is detected, the server replies based on the unmatched user reply statement; after replying based on the unmatched user reply statement, the current return visit clarifying statement is repeatedly output.

If the user answers a question, the robot answers to the user's current intent, and then repeats the current clarification session for the user to provide mission critical information. For example, the current call-back clarification word is "do you need to query for bills? "the user replies the sentence as" speak a laughing word for me ", then the robot recognizes the intention of" speak a laughing word ", can give an answer of" there is a mountain in the front, there is a temple in the mountain, there are small and still in the temple telling a story, i is not small and still ", then output the current return visit clarification word" do you need to inquire about the bill? ".

In the question-answer processing method in this embodiment, when a user reply statement that does not match the current revisit clarified statement is detected, a reply is performed based on the unmatched user reply statement, and after the reply is performed based on the unmatched user reply statement, the current revisit clarified statement is output, so that the processing efficiency of the user question can be improved under the condition that the user answers nothing, and the user experience can be improved.

In one embodiment, the question-answering processing method further includes:

and (a1) acquiring unresolved user question sentences.

Specifically, when the service quality evaluation value of the intelligent customer service is lower than a preset evaluation threshold value, the server acquires unresolved user question sentences within a preset time length. The preset time period may be a short time period, for example, the preset time period may be a time period less than 24 hours. The preset duration can be configured according to the requirement. The unresolved user question refers to a question posed by an unresolved user.

In this embodiment, the server may obtain the unresolved user question that appears within the preset duration for the preset number of times. The step (a1) may be before obtaining the historical conversation corresponding to the unresolved user question, or after performing a return visit to the user who provided the unresolved user question based on a return visit clarification technique through the user access channel.

Step (a2), acquiring an extension sentence corresponding to an unresolved user question; the expanded sentences are matched with the knowledge point categories corresponding to the unresolved user question sentences.

Specifically, the server obtains an extension sentence corresponding to an unresolved user question. The expanded sentence can be next to or a sentence. An expanded sentence is a sentence that matches the knowledge point category of the unresolved user question. That is, the unresolved user question and the corresponding expanded sentence belong to the same knowledge point category. Knowledge point categories can be used to characterize the category to which the question belongs.

In this embodiment, the manner for acquiring the category of the target knowledge point corresponding to the unresolved question of the user includes: clustering unresolved user question sentences to obtain unresolved cluster clusters; and determining the target knowledge point category corresponding to the unsolved cluster in a manual marking mode.

In this embodiment, obtaining an expanded sentence corresponding to an unresolved user question includes: clustering the unresolved user question to obtain an unresolved cluster when the knowledge point types corresponding to the unresolved user question are more than one; each unresolved cluster corresponds to a target knowledge point category; and acquiring the extension sentences corresponding to the unsolved cluster clusters.

And (a3) training the knowledge point classification model based on the unresolved question of the user, the expanded sentence and the corresponding knowledge point category label to obtain an updated knowledge point classification model.

And the knowledge point category label is used for representing the correct knowledge point category of the question of the user and the matched extension sentence. The target knowledge point category label refers to a manually set category label corresponding to an unresolved user question. The knowledge point classification model may be, specifically, a CNN (Convolutional Neural Networks) model or the like, but is not limited thereto. The knowledge point classification model is trained based on the question of the sample user and the corresponding sample knowledge point label.

Specifically, the server inputs unresolved user question sentences and corresponding knowledge point category labels, expanded sentences and corresponding target knowledge point category labels to the knowledge point classification model respectively for training until the training reaches a convergence condition, and then the updated knowledge point classification model is obtained. The convergence condition may be that the loss value is smaller than a preset loss value, the change of the model parameter between two iterations of the model is smaller than a preset change value, or the number of iterations reaches a preset number.

And (a4) performing question answering processing based on the updated knowledge point classification model.

Specifically, the server replaces the knowledge point classification model on the line with the updated knowledge point classification model. And the server receives the user question, inputs the user question into the updated knowledge point classification model to obtain a knowledge point category corresponding to the user question, and acquires a corresponding reply sentence based on the knowledge point category.

In the question-answer processing method, the unresolved user question and the corresponding expanded sentences within the preset time length are obtained, namely, the user question is subjected to sentence expansion of the same knowledge point category; and training the knowledge point classification model based on the unsolved user question, the expanded sentence and the corresponding knowledge point category label to obtain an updated knowledge point classification model, namely when hot spot problems in a short period are accumulated, directly training the knowledge point classification model based on the unsolved user question and the expanded question matched with the knowledge point category instead of training the knowledge point classification model by adopting all the user questions, and adopting the expanded sentence to improve the training efficiency of the knowledge point classification model and the accuracy of the knowledge point classification model at the same time.

In one embodiment, obtaining an expanded sentence corresponding to an unresolved question sentence includes: acquiring a historical user question set; acquiring knowledge point categories corresponding to all historical user question sets in a historical user question set; and searching a first historical user question matched with the target knowledge point category of the unresolved user question from the historical user question set based on the knowledge point categories corresponding to the historical user questions, and taking the first historical user question as an extension sentence of the unresolved user question.

The historical user question is a question at a time before the unresolved user question is acquired. The number of the historical user question in the historical user question set is not limited. The number of knowledge point categories corresponding to the multiple historical user question sentences may be multiple. Each historical user question has a corresponding knowledge point category. Plural means at least two. The target knowledge point category refers to a knowledge point category corresponding to an unresolved user question. And the knowledge point category corresponding to the first historical user question is the same as the target knowledge point category.

Specifically, the server obtains a plurality of historical user question sentences and knowledge point categories corresponding to the historical user question sentences. Since the historical user question is generally a question that has been subjected to question-answering processing, the knowledge point category corresponding to the historical user question can be directly acquired. And the server searches a first historical user question matched with the target knowledge point category of the unresolved user question from the historical user question set based on the knowledge point categories corresponding to the historical user questions. And the server takes the first historical user question as an expanded sentence corresponding to the unresolved user question.

Optionally, the server re-inputs the historical user question sets into the knowledge point classification model, and obtains the knowledge point categories corresponding to the historical user question sets.

In the question-answer processing method in this embodiment, a historical user question set and knowledge point categories corresponding to historical user questions are obtained, a first historical user question matched with a target knowledge point category of an unresolved user question is searched from the historical user question set based on the knowledge point categories corresponding to the historical user question and the knowledge point categories corresponding to the historical user question, the first historical user question is used as an extension sentence corresponding to the unresolved user question, that is, the historical user question of the same category is used as an extension sentence, and the obtained extension sentence is more relevant to an actual use scene, so that the identification accuracy of an updated knowledge point classification model is improved.

In one embodiment, searching a first historical user question from a set of historical user questions based on a knowledge point category corresponding to each historical user question, the first historical user question matching a target knowledge point category of an unresolved user question, includes: acquiring category probability values of all historical user question sentences corresponding to the categories of the knowledge points; and searching a first historical user question which is matched with the target knowledge point type of the unresolved user question and has the type probability higher than a first probability threshold value on the basis of the knowledge point type corresponding to each historical user question and the corresponding type probability value from the historical user question set.

The category probability value refers to the category probability that a question of a historical user belongs to a certain knowledge point category. For example, the category probability that a historical user question belongs to knowledge point category a is 95%. The first probability threshold may be set according to actual demand. For example, the first probability threshold may be 95%, 90%, 85%, etc., but is not limited thereto.

Specifically, the server acquires class probability values corresponding to the knowledge point classes of the historical user questions, searches the historical user questions matched with the target knowledge point classes of the unresolved user questions from the historical user question set based on the knowledge points corresponding to the historical user questions and the corresponding class probability values, and screens out first historical user questions from the historical user questions matched with the target knowledge point classes based on the class probability values corresponding to the target knowledge point classes, wherein the class probability of the target knowledge point classes is higher than a first probability threshold value.

The question-answer processing method in this embodiment obtains category probability values corresponding to the categories of knowledge points, searches, based on the categories of knowledge points corresponding to the historical user question sentences and the corresponding category probability values, questions matched with the target knowledge point categories of the unresolved user question sentences from the historical user question sentence set, that is, finds questions matched with the categories, where the category probability of the target knowledge point categories is higher than a first probability threshold, that is, the historical user question sentences with higher category credibility are obtained, so that the accuracy of the obtained expanded question sentences is higher, thereby improving the classification accuracy of the updated knowledge point classification model, and because fewer historical user questions satisfying the first probability threshold, the training duration of the knowledge point classification model can also be reduced.

In one embodiment, obtaining an expanded sentence corresponding to an unresolved user question includes: acquiring session information corresponding to unresolved user question sentences; acquiring a word slot from the session information; and taking the word slot as an expansion word corresponding to the unresolved user question.

Specifically, the server obtains session information corresponding to unresolved user question sentences. The session information refers to a session sentence obtained in a session corresponding to an unresolved user question. The session information in the unresolved user question may be acquired from a session interface of the unresolved user question, or may be acquired from voice data corresponding to the unresolved user question. And the server acquires the word slot from the session information and takes the word slot as an expansion word corresponding to the unresolved question of the user.

In the question-answer processing method in this embodiment, the word slot is obtained from the session information by obtaining the session information corresponding to the unresolved user question, and the word slot is used as the expanded word corresponding to the unresolved user question, so that the unresolved user question can be expanded, training data is expanded, and the updated knowledge point classification model is more accurate.

In one embodiment, before obtaining the unresolved user question, the question-answer processing method further includes: acquiring category probability values of all historical user question sets belonging to all knowledge points; when a second historical user question exists in the historical user question set based on the category probability value of each historical user question belonging to each knowledge point category, clustering the second historical user question to obtain a cluster; the second historical user question is a question with the maximum category probability smaller than a second probability threshold; and training the original knowledge point classification model based on the clustering clusters and the corresponding clustering class labels, and obtaining the knowledge point classification model when the training reaches the preset condition.

The historical user question refers to a user question acquired before an unresolved user question is acquired. Each historical user question has a corresponding knowledge point category. And more than one knowledge point category corresponding to one historical user question can be provided. The second probability threshold may be set as desired. And the second probability threshold is less than the first probability threshold. The second probability threshold is used to characterize a smaller probability, for example, the second probability threshold may be 50%, 40%, etc. without limitation. The second historical user question is a question with a maximum category probability less than a second probability threshold. The cluster comprises at least one second historical user question. And each cluster corresponds to a knowledge point category. Namely, each cluster has a corresponding cluster category label. The second historical question sentences in the same cluster have high similarity. And the clustering class labels are used for representing the knowledge point classes corresponding to the clustering clusters. The different clustering clusters have different corresponding knowledge point categories and different corresponding clustering category labels. The clustering method is, for example, a kfmeans (k-means) text clustering method based on tf-idf (term frequency-inverse text frequency index).

The preset condition may refer to, but not limited to, that the number of training times reaches a preset number of times, or that the output class probability value is greater than a preset probability value. The original knowledge point classification model is a classification model which has a knowledge point classification function and is not subjected to clustering training of second historical question sentences.

Specifically, since the historical user question is generally a question that has been subjected to question-answering processing, the category probability values of the historical user question belonging to the respective knowledge points can be directly obtained. And when the question with the maximum class probability smaller than the second probability threshold exists in the historical user question set based on the class probability values belonging to the classes of the knowledge points, clustering the second historical user question to obtain a cluster. The server may add the cluster to a knowledge base. And training the original knowledge point classification model based on the clustering clusters and the corresponding clustering class labels, and obtaining the knowledge point classification model when the training reaches the preset condition. For example, if the second probability threshold is 40%, the probability that a question of a historical user belongs to knowledge point a is 20%, the probability that the question belongs to knowledge point B is 30%, the probability that the question belongs to knowledge point C is 35%, and the probability that the question belongs to knowledge point D is 15%, the maximum category probability is determined to be 35%, and the maximum probability 35% is less than the second probability threshold 40%, then the question of the historical user is the second historical user question. And clustering the second historical user question to obtain a cluster.

In the question-answer processing method in this embodiment, category probability values of the respective historical user question sets belonging to the respective knowledge point categories are obtained, and when it is determined that second historical user question sets exist in the historical user question sets based on the category probability values of the respective historical user question sets belonging to the respective knowledge point categories, it is described that the second historical question sets can belong to new knowledge points, and therefore, clustering processing needs to be performed on the second historical user question sets to obtain a cluster; training an original knowledge point classification model based on the clustering clusters and the corresponding clustering category labels, obtaining the knowledge point classification model when the training reaches a preset condition, namely screening out a less accurate historical user question, namely a second historical question, and performing retraining to obtain the knowledge point classification model which can classify more knowledge point categories and is more accurate in classification.

In one embodiment, obtaining an unresolved user question includes: acquiring a user question and a user behavior corresponding to the user question; unresolved user question is determined based on user behavior.

Acquiring an expanded sentence corresponding to an unresolved user question, wherein the expanded sentence comprises the following steps: acquiring various evaluation index values of the intelligent customer service; determining a service quality evaluation value of the intelligent customer service based on each evaluation index value; and when the service quality evaluation value is lower than a preset evaluation threshold value, acquiring an expanded sentence corresponding to the unresolved user question sentence.

Wherein the user behavior is used to characterize the user's evaluation of unresolved user question sentences. The user behavior can be specifically unresolved, manual, task midway hang-up, praise, step on, normally end conversation and other behaviors. And the user question sentence is an unresolved user question sentence which is indicated by unresolved question, manual work change, suspension of the task in the middle and point stepping. And if the conversation is complied with, normally ending the conversation, the user question is a resolved user question.

The intelligent customer service refers to a robot for providing question and answer processing for users. The intelligent customer service obtains the user question and responds based on the user question to obtain the reply sentence.

The evaluation index is an index which is set for the intelligent customer service by the server and is used for evaluating the service quality of the intelligent customer service. The evaluation index value comprises an index value generated by evaluating the intelligent customer service by the user. The evaluation index value may further include an index value obtained by counting up user questions and unresolved user questions over a period of time. And the indexes corresponding to different user evaluations may be different. For example, the number of user likes or clicks is used as the index a, the solved problem and the unsolved problem are used as the index B, and the resolution is not limited thereto as the index C.

Specifically, the server obtains a user question within a preset time length and a user behavior corresponding to the user question. The server determines an unresolved user question based on user behavior within a preset duration. The server obtains various evaluation index values of the user to the intelligent customer service in a preset time. The server counts each evaluation index value, obtains the weight corresponding to each evaluation index value, and determines the service quality evaluation value of the intelligent customer service based on each index value and the corresponding weight. And when the service quality evaluation value in the preset time is lower than a preset evaluation threshold value, acquiring an expanded sentence corresponding to the unresolved user question sentence. For example, the server obtains a user question Z, and the user behavior corresponding to the user question Z is "click-on", so that the user question Z is an unresolved user question. The server obtains the evaluation of the user on the intelligent customer service in a period of time, such as service satisfaction degree scoring, approval or stepping. Manual operation and the like, and counting evaluation index values in a period of time, such as the number of solved/unsolved problems, the interception rate, the solution rate and the like, so as to determine the service quality evaluation value of the intelligent customer service.

In the question-answer processing method in this embodiment, a user question and a user behavior corresponding to the user question are acquired, and an unresolved user question is determined based on the user behavior, so that the unresolved user question is determined; acquiring various evaluation index values of the intelligent customer service, and determining a service quality evaluation value of the intelligent customer service based on the various index values, so that a quality evaluation value of the knowledge point classification model can be obtained; when the service quality evaluation value is lower than the preset evaluation threshold value, it is described that the accuracy of the knowledge point classification model is not high, and the problem proposed by the user cannot be solved, so that an expanded sentence corresponding to an unresolved user question needs to be acquired, and the knowledge point classification model is subjected to emergency training to relieve the service pressure caused in a short term.

In one embodiment, a question-answer processing method includes:

and (b1) acquiring the category probability value of each historical user question in the historical user question set belonging to each knowledge point category.

And (b2) clustering the second historical user question to obtain a cluster when the second historical user question exists in the historical user question set based on the category probability value of each historical user question belonging to each knowledge point category. The second historical user question is a question with a maximum category probability less than a second probability threshold.

And (b3) training the original knowledge point classification model based on the clustering clusters and the corresponding clustering class labels, and obtaining the knowledge point classification model when the training reaches the preset condition.

And (b4) acquiring the user question and the user behavior corresponding to the user question.

Step (b5), determining unresolved user question based on the user behavior.

And (b6) acquiring various evaluation index values of the intelligent customer service.

And a step (b7) of determining a service quality evaluation value of the intelligent customer service based on each evaluation index value.

And (b8) acquiring a historical user question set when the service quality evaluation value is lower than a preset evaluation threshold value.

And (b9) acquiring the knowledge point types corresponding to the historical user question in the historical user question set.

And (b10) acquiring the category probability value corresponding to the knowledge point category of each historical user question.

And (b11) searching a first historical user question which is matched with the target knowledge point type of the unresolved user question and has the category probability higher than a first probability threshold value on the basis of the knowledge point type corresponding to each historical user question and the corresponding category probability value.

And (b12) taking the first historical user question as an expanded sentence corresponding to the unresolved user question. The expanded sentences are matched with the knowledge point categories corresponding to the unresolved user question sentences.

And (b13) acquiring the session information corresponding to the unresolved user question.

And step (b14), acquiring the word slot from the conversation information.

And (b15) taking the word slot as an expansion word corresponding to the unresolved user question.

And (b16) training the knowledge point classification model based on the unresolved user question, the expanded sentence and the corresponding knowledge point category label to obtain an updated knowledge point classification model.

And (b17) performing question answering processing based on the updated knowledge point classification model.

And (b18) acquiring the historical conversation corresponding to the unresolved user question.

And (b19) determining the corresponding task flow according to the intention of the historical conversation.

And (b20) determining the return visit clarifying words corresponding to the unresolved user question according to the task flow.

And (b21) acquiring a user access channel corresponding to the unresolved user question.

And (b22) revisiting the user who provides the unresolved user question through the user access channel based on the revisiting clarification technique to complete the task flow.

In the question-answer processing method in this embodiment, when hot problems in a short period are accumulated, the knowledge point classification model is trained directly based on the unresolved user question and the extended question matched with the category of the knowledge point, instead of the method of training the knowledge point classification model by using all the user questions, and the extended sentence is used, so that the accuracy of the knowledge point classification model can be improved while the training efficiency of the knowledge point classification model is improved, and the question-answer processing is performed based on the updated knowledge point classification model, that is, the unresolved user question can be replied based on the updated knowledge point classification model, so that the manual customer service pressure is relieved, and the user is helped to solve the unresolved problems by revising the user, and the user experience is provided.

In one embodiment, in the conversation process, the robot answers questions of the user frequently, the user frequently turns to manual work or hangs up or complaints, the manual customer service volume is increased suddenly, the customer service staff is tired of answering the questions, and the problem is not solved by the user and accumulated, which may be caused by the fuzzy intention of the user due to unclear description or the low recognition accuracy of a certain intention or word slot of the robot on certain nodes of certain tasks. If the problem is a task with a high access heat, the influence of the problem is continuously expanded, and a large brand loss is caused to enterprises. In the prior art, an intelligent customer service operator obtains the quality of the intelligent customer service through scoring service satisfaction or ' praise/step on ' solved/unsolved ' interception rate ' and ' solution rate ' of a user, and then adjusts a task flow and optimizes the robot effect according to the statistic conditions of indexes of ' step on ' unsolved ', ' low grade ', ' low interception rate ' and ' low solution rate ' of a task. When the user visits the robot again, the intelligent customer service can accurately identify the user intention and the word slot, and help to solve the problem or handle the business. However, as the service quality of the robot is reduced to be improved, the following stages need to be passed in the middle: the service quality of the robot is reduced → the operator knows that the service quality of the robot is low → the optimization of the knowledge base is completed, which takes a long time and accumulates a large number of unsolved problems. During this time, manual customer service is under greater service pressure and the user may give up on resolving the problem by waiting too long for manual customer service. On the other hand, the user cannot know the optimized information of the intelligent customer service, and the possibility of accessing the intelligent customer service again to solve the problem is low.

Therefore, the intelligent customer service exception handling mechanism is provided, a large number of unsolved problems caused by poor service quality of the short-term robot can be quickly handled through the mechanism, the pressure of manual customer service is relieved, users are actively helped to solve the problems, and user experience is enhanced. As shown in fig. 3, a schematic flow chart of a question-answering processing method in another embodiment includes:

step 302, the user accesses the intelligent customer service, and the intelligent customer service replies to the user question.

Specifically, the user accesses a telephone or a text to access the intelligent customer service, and the intelligent customer service replies to the user question.

Step 304, the intelligent customer service automatically identifies unresolved question sentences.

Specifically, the unsolved problem is obtained through user behaviors such as "unsolved", "manual", "hanging up in the middle of a task", and the like, and the user access channel information is recorded.

And step 306, carrying out statistical analysis on the service quality evaluation value of the intelligent customer service.

Specifically, the robot summarizes service satisfaction degree scores or numbers of ' praise/step on ', numbers of ' solved/unsolved problems ', interception rates ' and ' solution rates ' information, and combines the information to give a service quality evaluation value of the intelligent customer service. The service satisfaction degree score or the number of 'praise/step on', the number of 'solved/unsolved problems', the interception rate 'and the solution rate' can be all evaluation index values.

Step 308, an alarm of low quality of service is issued.

Specifically, when the service evaluation quality value of the intelligent customer service is lower than a preset evaluation threshold value, an alarm with low service quality is sent out.

Step 310, optimizing the knowledge base.

Specifically, when the service evaluation quality value of the intelligent customer service is lower than a preset evaluation threshold, an expanded sentence corresponding to an unresolved user question sentence is obtained. And the server trains the knowledge point classification model based on the unresolved user question, the expanded sentence and the corresponding knowledge point class label to obtain an updated knowledge point classification model.

Optionally, acquiring a category probability value of each historical user question belonging to each knowledge point category in the historical user question set; when a second historical user question exists in the historical user question set based on the category of each historical user question belonging to each knowledge point category, clustering the second historical user question to obtain a cluster; the second historical user question is a question with the maximum category probability smaller than a second probability threshold; and training the knowledge point classification model based on the unresolved question of the user, the expanded sentence, the corresponding knowledge point category label, the clustering cluster and the corresponding clustering category label, and obtaining the updated knowledge point classification model when the training reaches a preset condition.

Step 312, the intelligent customer service return visit mechanism is started.

Specifically, the intelligent customer service return visit mechanism acquires a user access channel corresponding to an unresolved user question; and through the user access channel, the user who provides the unresolved user question mark is revisited, and a message related to the unresolved user question is sent to the user so as to solve the problem in the unresolved user question. And the robot return visit function actively returns the recorded unsolved user problems to the user through a user access channel, and if the user calls to access the intelligent customer service robot, the intelligent customer service robot dials the user call. The robot revisiting function changes passiveness into initiative, a fixed telephone is generated according to a template, and a user is actively asked for a problem, such as 'you are your intelligent customer service robot small w of a certain company, i realize upgrade optimization, and you consult a certain problem before and ask me to serve your bar again'. After the robot actively asks the user problems, the task flow chart corresponding to a certain problem is continuously executed, the user is guided to provide word slots and intention information through clarification, the whole task processing flow is completed, and the user is helped to solve the problems. On one hand, the updated and optimized information of the robot can be informed to the user, so that the confidence of the user on the intelligent customer service capability is improved; on the one hand, the system can actively help to solve the user problem.

The question-answer processing method in the embodiment can improve the overall service quality of the intelligent customer service, even if the short-term service quality is reduced, the unsolved user problems can be resolved under the robot return visit mechanism, the overall solution rate is improved, the manual customer service pressure is effectively relieved, the active return visit user helps solve the problems, and the user experience can be effectively enhanced.

It should be understood that although the steps in the flowcharts of fig. 2 and 3 are shown in order as indicated by the arrows, the steps are not necessarily performed in order as indicated by the arrows. The steps are not performed in the exact order shown and described, and may be performed in other orders, unless explicitly stated otherwise. Moreover, at least some of the steps in fig. 2 and 3 may include multiple steps or multiple stages, which are not necessarily performed at the same time, but may be performed at different times, which are not necessarily performed in sequence, but may be performed in turn or alternately with other steps or at least some of the other steps.

In one embodiment, as shown in fig. 4, there is provided a question and answer processing apparatus including: a historical session acquisition module 402, a task flow determination module 404, a clarification session determination 406, and a return visit module 408, wherein:

a history session obtaining module 402, configured to obtain a history session corresponding to an unresolved question of a user;

a task flow determining module 404, configured to determine a corresponding task flow according to an intention of the historical session;

a clarification word operation determining module 406, configured to determine a return visit clarification word operation corresponding to the unresolved user question according to the task flow;

a return access module 408, configured to obtain a user access channel corresponding to an unresolved user question;

the revisiting module 408 is further configured to revisit, through the user access channel, the user who presents the unresolved user question based on a revisiting clarification technique to complete the task flow.

The question-answer processing device in the embodiment can obtain more accurate return visit clarification dialogs by obtaining the history conversation corresponding to the unresolved user question, determining the corresponding task flow according to the intention of the history conversation, and determining the return visit clarification dialogs according to the task flow; the method comprises the steps of obtaining a user access channel corresponding to an unresolved user question, and actively returning to the unresolved robot through the user access channel, so that the confidence of a user on intelligent customer service capacity can be improved, the problem of the user can be actively solved, customer service pressure is relieved, and user experience is improved.

In one embodiment, the revisit module 408 is further configured to obtain a user reply statement obtained by revisit clarification session through the user access channel; extracting word slots and intention information from the user reply sentences; performing information query according to the word slot and the intention information to obtain an information query result; and returning an information query result through the user access channel.

The question-answering processing device in the embodiment acquires the user reply sentences obtained by revisiting the clarified dialogs, extracts the word slots and the intention information from the user reply sentences, performs information query according to the word slots and the intention information to obtain information query results, and returns the information query results through the user access channel, so that good communication with the user can be established, the user problem can be actively solved, and the user experience is improved.

In one embodiment, the revisit module 408 is further configured to obtain a maximum number of revisits for a case where the user does not reply; and when the number of times of visiting the user back through the user access channel reaches the maximum number of times of visiting back, marking the unresolved user problem, and suspending processing of the unresolved user question.

The question-answer processing device in this embodiment obtains the maximum number of revisits to the situation where the user does not reply, marks unresolved user question when the number of revisits the user through the user access channel reaches the maximum number of revisits, and suspends processing of the unresolved user question to reduce resource waste.

In one embodiment, the return access module 408 is further configured to return access to the user who provided the unresolved user question based on a return access clarification technique during the user idle period through the user access channel to complete the task flow.

The question-answer processing device in the embodiment performs the return visit to the user who provides the unresolved user question based on the return visit clarification technique to complete the task flow in the user idle time period through the user access channel, so that the convenience and the possibility of answering by the user can be improved, and the unresolved user question can be solved.

In one embodiment, the revisit module 408 is further configured to obtain a corresponding mood-relieving word when the user mood is identified as a preset bad mood;

The question-answer processing device in the embodiment acquires corresponding words of the relaxed emotion when recognizing that the emotion of the user is a preset poor emotion, performs return visit to the user who presents unresolved user question through the user access channel based on the return visit clarifying words after outputting the words of the relaxed emotion to complete the task flow, and can improve the processing efficiency of the user question and improve the user experience under the conditions that the user is not matched or abused.

In one embodiment, the revisiting module 408 is further configured to, when a user reply statement that does not match the current revisiting clarification session is detected, reply based on the non-matching user reply statement; after replying based on the unmatched user reply statement, the current return visit clarifying statement is output.

The question-answering processing device in the embodiment can improve the processing efficiency of the question of the user and improve the user experience under the condition that the user answers no question by outputting the current return visit clarifying question after replying based on the unmatched user reply sentence when detecting the user reply sentence unmatched with the current return visit clarifying question.

In one embodiment, the question-answering processing method further includes a first obtaining module, a second obtaining module, a training module and a processing module, wherein:

the first acquisition module is used for acquiring unresolved user question sentences;

the second acquisition module is used for acquiring the expanded sentences corresponding to the unresolved user question sentences; the unresolved question of the user is matched with the knowledge point category corresponding to the expanded statement;

the training module is used for training the knowledge point classification model based on the unresolved question and expanded sentence of the user and the corresponding knowledge point category label to obtain an updated knowledge point classification model;

and the processing module is used for performing question answering processing based on the updated knowledge point classification model.

The question-answer processing device obtains the unresolved user question and the corresponding expanded sentence, namely, the user question is subjected to sentence expansion of the same knowledge point type; and training the knowledge point classification model based on the unsolved user question, the expanded sentence and the corresponding knowledge point category label to obtain an updated knowledge point classification model, namely when hot spot problems in a short period are accumulated, directly training the knowledge point classification model based on the unsolved user question and the expanded question matched with the knowledge point category instead of training the knowledge point classification model by adopting all the user questions, and adopting the expanded sentence to improve the training efficiency of the knowledge point classification model and the accuracy of the knowledge point classification model at the same time.

In one embodiment, the second obtaining module is used for obtaining a historical user question set; acquiring knowledge point categories corresponding to all historical user question sets in a historical user question set; and searching a first historical user question matched with the target knowledge point category of the unresolved user question from the historical user question set based on the knowledge point categories corresponding to the historical user questions, and taking the first historical user question as an extension sentence of the unresolved user question.

The question-answer processing device in this embodiment obtains a historical user question set and knowledge point categories corresponding to historical user questions, searches for a first historical user question matched with a target knowledge point category of an unresolved user question from the historical user question set based on the knowledge point categories corresponding to the historical user question and based on the knowledge point categories corresponding to the historical user question, takes the first historical user question as an extension sentence corresponding to the unresolved user question, that is, takes the historical user question of the same category as the extension sentence, and makes the obtained extension sentence more appropriate for an actual use scene, thereby improving the identification accuracy of the updated knowledge point classification model.

In one embodiment, the second obtaining module is used for obtaining a category probability value corresponding to a knowledge point category of each historical user question; and searching a first historical user question which is matched with the target knowledge point type of the unresolved user question and has the type probability higher than a first probability threshold value on the basis of the knowledge point type corresponding to each historical user question and the corresponding type probability value from the historical user question set.

The question-answer processing device in this embodiment obtains the category probability value corresponding to the knowledge point category, and based on the knowledge point category corresponding to the historical user question and the corresponding category probability value, finds out a question matched with the category from the historical user question set, where the category probability of the target knowledge point category is higher than the first probability threshold, that is, the historical user question with higher category credibility is obtained, so that the accuracy of the obtained expanded question is higher, thereby improving the classification accuracy of the updated knowledge point classification model, and because fewer historical user questions satisfying the first probability threshold, the training duration of the knowledge point classification model can also be reduced.

In one embodiment, the second obtaining module is configured to obtain session information corresponding to unresolved user question sentences; acquiring a word slot from the session information; and taking the word slot as an expansion word corresponding to the unresolved user question.

The question-answering processing device in this embodiment can expand the unresolved user question by acquiring the session information corresponding to the unresolved user question, acquiring the word slot from the session information, and using the word slot as the expanded word corresponding to the unresolved user question, thereby expanding the training data and making the updated knowledge point classification model more accurate.

In one embodiment, the training module is further configured to obtain category probability values of knowledge points to which each historical user question belongs in the historical user question set; when a second historical user question exists in the historical user question set based on the category probability value of each historical user question belonging to each knowledge point category, clustering the second historical user question to obtain a cluster; the second historical user question is a question with the maximum category probability smaller than a second probability threshold; and training the original knowledge point classification model based on the clustering clusters and the corresponding clustering class labels, and obtaining the knowledge point classification model when the training reaches the preset condition.

The question-answer processing device in the embodiment acquires class probability values of all historical user question sets belonging to all knowledge point classes, and when second historical user question sets exist in the historical user question sets based on the class probability values of all the historical user question sets belonging to all the knowledge point classes, the second historical user question sets can belong to new knowledge points, so that the second historical user question sets need to be clustered to obtain a cluster; training an original knowledge point classification model based on the clustering clusters and the corresponding clustering category labels, obtaining the knowledge point classification model when the training reaches a preset condition, namely screening out a less accurate historical user question, namely a second historical question, and performing retraining to obtain the knowledge point classification model which can classify more knowledge point categories and is more accurate in classification.

In one embodiment, the first obtaining module is configured to obtain a user question and a user behavior corresponding to the user question; unresolved user question is determined based on user behavior. The second acquisition module is used for acquiring various evaluation index values of the intelligent customer service; determining a service quality evaluation value of the intelligent customer service based on each evaluation index value; and when the service quality evaluation value is lower than a preset evaluation threshold value, acquiring an expanded sentence corresponding to the unresolved user question sentence.

The question-answer processing device in this embodiment acquires a user question and a user behavior corresponding to the user question, and determines an unresolved user question based on the user behavior, thereby determining the unresolved user question; acquiring various evaluation index values of the intelligent customer service, and determining a service quality evaluation value of the intelligent customer service based on the various index values, so that a quality evaluation value of the knowledge point classification model can be obtained; when the service quality evaluation value is lower than the preset evaluation threshold value, it is described that the accuracy of the knowledge point classification model is not high, and the problem proposed by the user cannot be solved, so that an expanded sentence corresponding to an unresolved user question needs to be acquired, and the knowledge point classification model is subjected to emergency training to relieve the service pressure caused in a short term.

For specific limitations of the question-answering processing device, reference may be made to the above limitations of the question-answering processing method, which are not described herein again. The modules in the above-mentioned question and answer processing device can be wholly or partially implemented by software, hardware and a combination thereof. The modules can be embedded in a hardware form or independent from a processor in the computer device, and can also be stored in a memory in the computer device in a software form, so that the processor can call and execute operations corresponding to the modules.

In one embodiment, a computer device is provided, which may be a server, the internal structure of which may be as shown in fig. 5. The computer device includes a processor, a memory, and a network interface connected by a system bus. Wherein the processor of the computer device is configured to provide computing and control capabilities. The memory of the computer device comprises a nonvolatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of an operating system and computer programs in the non-volatile storage medium. The database of the computer device is used for storing statement data. The network interface of the computer device is used for communicating with an external terminal through a network connection. The computer program is executed by a processor to implement a question-answering processing method.

Those skilled in the art will appreciate that the architecture shown in fig. 5 is merely a block diagram of some of the structures associated with the disclosed aspects and is not intended to limit the computing devices to which the disclosed aspects apply, as particular computing devices may include more or less components than those shown, or may combine certain components, or have a different arrangement of components.

In one embodiment, a computer device is further provided, which includes a memory and a processor, the memory stores a computer program, and the processor implements the steps of the above method embodiments when executing the computer program.

In an embodiment, a computer-readable storage medium is provided, in which a computer program is stored which, when being executed by a processor, carries out the steps of the above-mentioned method embodiments.

In one embodiment, a computer program product or computer program is provided that includes computer instructions stored in a computer-readable storage medium. The computer instructions are read by a processor of a computer device from a computer-readable storage medium, and the computer instructions are executed by the processor to cause the computer device to perform the steps in the above-mentioned method embodiments.

It will be understood by those skilled in the art that all or part of the processes of the methods of the embodiments described above can be implemented by hardware instructions of a computer program, which can be stored in a non-volatile computer-readable storage medium, and when executed, can include the processes of the embodiments of the methods described above. Any reference to memory, storage, database or other medium used in the embodiments provided herein can include at least one of non-volatile and volatile memory. Non-volatile Memory may include Read-Only Memory (ROM), magnetic tape, floppy disk, flash Memory, optical storage, or the like. Volatile Memory can include Random Access Memory (RAM) or external cache Memory. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM), among others.

The technical features of the above embodiments can be arbitrarily combined, and for the sake of brevity, all possible combinations of the technical features in the above embodiments are not described, but should be considered as the scope of the present specification as long as there is no contradiction between the combinations of the technical features.

The above-mentioned embodiments only express several embodiments of the present application, and the description thereof is more specific and detailed, but not construed as limiting the scope of the invention. It should be noted that, for a person skilled in the art, several variations and modifications can be made without departing from the concept of the present application, which falls within the scope of protection of the present application. Therefore, the protection scope of the present patent shall be subject to the appended claims.

Claims

1. A question-answer processing method, characterized in that the method comprises:

acquiring a user access channel corresponding to the unresolved user question;

2. The method of claim 1, further comprising:

acquiring a user reply sentence obtained by revisiting a clarifying dialog;

extracting word slots and intention information from the user reply sentences;

performing information query according to the word slot and the intention information to obtain an information query result;

and returning the information query result through the user access channel.

3. The method of claim 1, further comprising:

acquiring the maximum return visit times of the user on the condition of not replying;

and when the number of times of visiting back the user through the user access channel reaches the maximum number of times of visiting back, marking the unresolved user question, and suspending processing of the unresolved user question.

4. The method of claim 1, wherein the revisiting, through the user access channel, the user who presented the unresolved user question based on the revisiting clarification technique to complete the task flow comprises:

and returning the user who provides the unresolved user question based on the return visit clarification word in the idle time period of the user through the user access channel so as to complete the task flow.

5. The method of claim 1, further comprising:

when the emotion of the user is recognized to be a preset poor emotion, corresponding words of relieving the emotion are obtained;

and after outputting the words of the relieved emotion, executing the revisit of the user who proposes the unresolved user question through the user access channel based on the revisit clarification words so as to complete the task flow.

6. The method of claim 1, further comprising:

when a user reply statement which is not matched with the current return visit clarification word is detected, replying based on the unmatched user reply statement;

after replying based on the unmatched user reply statement, outputting the current return visit clarifying statement.

7. The method of claim 1, further comprising:

acquiring an expanded sentence corresponding to the unresolved user question; the expanded sentences are matched with the knowledge point categories corresponding to the unresolved user question sentences;

training a knowledge point classification model based on the unresolved user question, the expanded sentence and the corresponding knowledge point category label to obtain an updated knowledge point classification model;

and performing question answering processing based on the updated knowledge point classification model.

8. The method of claim 7, wherein the obtaining the expanded sentence corresponding to the unresolved user question sentence comprises:

acquiring a historical user question set;

acquiring knowledge point categories corresponding to all historical user question sentences in the historical user question sentence set;

and searching a first historical user question matched with the target knowledge point category of the unresolved user question from the historical user question set based on the knowledge point category corresponding to each historical user question, and taking the first historical user question as an expanded sentence corresponding to the unresolved user question.

9. The method of claim 8, wherein the searching for the first historical user question from the set of historical user questions that matches the target knowledge point category of the unresolved user question based on the knowledge point category corresponding to each of the historical user questions comprises:

acquiring category probability values of the historical user question sentences corresponding to the categories of the knowledge points;

and searching a first historical user question which is matched with the target knowledge point category of the unresolved user question and has the category probability higher than a first probability threshold value from the historical user question set on the basis of the knowledge point category corresponding to each historical user question and the corresponding category probability value.

10. The method of claim 7, wherein the obtaining the expanded sentence corresponding to the unresolved user question sentence comprises:

acquiring session information corresponding to the unresolved user question;

acquiring a word slot from the session information;

and taking the word slot as an expansion word corresponding to the unresolved user question.

11. The method of claim 7, further comprising:

acquiring category probability values of all historical user question sets belonging to all knowledge point categories;

when a second historical user question exists in the historical user question set based on the category probability value of each historical user question belonging to each knowledge point category, clustering the second historical user question to obtain a cluster; the second historical user question is a question with the maximum category probability smaller than a second probability threshold;

and training an original knowledge point classification model based on the clustering clusters and the corresponding clustering class labels, and obtaining the knowledge point classification model when the training reaches a preset condition.

12. The method according to any one of claims 7 to 11, wherein the obtaining of the unresolved user question comprises:

acquiring a user question and a user behavior corresponding to the user question;

determining an unresolved user question based on the user behavior;

the obtaining of the expanded sentence corresponding to the unresolved user question sentence includes:

acquiring various evaluation index values of the intelligent customer service;

determining a service quality evaluation value of the intelligent customer service based on various evaluation index values;

and when the service quality evaluation value is lower than a preset evaluation threshold value, acquiring an expanded sentence corresponding to the unresolved user question sentence.

13. A question-answering processing apparatus characterized by comprising:

14. A computer device comprising a memory and a processor, the memory storing a computer program, characterized in that the processor, when executing the computer program, implements the steps of the method of any of claims 1 to 12.

15. A computer-readable storage medium, on which a computer program is stored, which, when being executed by a processor, carries out the steps of the method of any one of claims 1 to 12.