Problem expansion method and device, server and medium

Abstract

The application relates to the field of automatic programming, and discloses a problem expansion method and device, a server and a medium. The method comprises the following steps: acquiring an interaction flow of a user and a device, wherein the interaction flow comprises a question and an answer corresponding to the question; determining an effective question in the question, grouping the effective question according to the answer corresponding to the effective question, and determining a candidate content set; for each candidate content set, clustering the effective questions in the candidate content set to obtain a clustering result set; determining a target question set according to the clustering result set, and adding the target question set into a preset question database to expand the questions in the preset question database. The technical scheme of the embodiment of the application solves the problem that an effective answer cannot be obtained when a user interacts with an interactive device, improves the business question material collection efficiency of a customer service robot, effectively supplements a question library, reduces the manual maintenance cost, and improves the question and answer matching rate effect.

Description

TECHNICAL FIELD

[0001] Embodiments of the present application relate to the field of automatic programming, and in particular to a problem expansion method and device, a server and a medium. BACKGROUND

[0002] With the rapid development of computer and artificial intelligence technology, many enterprises use intelligent customer service robots to replace manual customer service to reduce labor costs.

[0003] In the use scenario of intelligent customer service robots in a specific field, since the problem content requested by different customer groups usually has the characteristics of being short and highly interfering, the user sometimes needs to input the request content multiple times to obtain the relevant answer from the existing problem corpus, so it is necessary to expand the problem library of the customer service robot.

[0004] The prior art can only obtain the context relationship of the problem in the user dialogue, and there is great limitation for the expansion of the problem. SUMMARY

[0005] Embodiments of the present application provide a problem expansion method, device, server and medium to improve the efficiency of collecting business problem materials of the customer service robot, effectively supplement the problem library, reduce the manual maintenance cost, and improve the question and answer matching rate.

[0006] In a first aspect, the embodiments of the present application provide a problem expansion method, which comprises: obtaining an interaction flow of a user and a device, the interaction flow comprising a question and an answer corresponding to the question; determining an effective question in the question, grouping the effective question according to the answer corresponding to the effective question, and determining a candidate content set; for each candidate content set, clustering the effective questions in the candidate content set to obtain a clustering result set; determining a target question set according to the clustering result set, and adding the target question set to a preset problem database to expand the problems in the preset problem database.

[0007] In a second aspect, the embodiments of the present application also provide a problem expansion device, which comprises:

[0008] An interaction flow acquisition module is configured to obtain an interaction flow of a current user and a smart device, the interaction flow comprising a question and an answer corresponding to the question; a candidate content set determination module is configured to determine an effective question in the question, group the effective question according to the answer corresponding to the effective question, and determine a candidate content set; a clustering result set determination module is configured to cluster the effective questions for each candidate content set to obtain a clustering result set; and a problem expansion module is configured to determine a target question set according to the clustering result set, and add the target question set to a preset problem database to expand the problems in the database.

[0009] In a third aspect, an embodiment of the present application further provides a server, wherein the server comprises:

[0010] one or more processors;

[0011] a storage device configured to store one or more programs;

[0012] When the one or more programs are executed by the one or more processors, the one or more processors implement the problem expansion method provided by any embodiment of the present application.

[0013] In a fourth aspect, an embodiment of the present application further provides a computer readable storage medium having a computer program stored thereon, wherein the program, when executed by a processor, implements the problem expansion method provided by any embodiment of the present application.

[0014] The technical scheme of the embodiment of the present application acquires the interaction flow of the user and the device, the interaction flow comprises a question and an answer corresponding to the question; determines the effective question in the question, groups the effective question according to the answer corresponding to the effective question, and determines a candidate content set; removes the invalid question in the question, improves the effectiveness of the question, and reduces the data processing pressure. For each candidate content set, the effective question in the candidate content set is clustered to obtain a clustering result set; questions with the same semantics are clustered together, which facilitates the expansion of the questions of the semantics; the target question set is determined according to the clustering result set, and the target question set is added to a preset question database, so as to expand the questions in the preset question database, solve the problem that an effective answer cannot be obtained when the user interacts with the interactive device, improve the business question material collection efficiency of the customer service robot, effectively supplement the question library, reduce the manual maintenance cost, and improve the question and answer matching rate effect. BRIEF DESCRIPTION OF DRAWINGS

[0015] Figure 1 is a flowchart of a problem expansion method in the embodiment one of the present application;

[0016] Figure 2 is a flowchart of effective request determination in the embodiment two of the present application;

[0017] Figure 3 is a flowchart of effective question grouping in the embodiment two of the present application;

[0018] Figure 4 is a flowchart of effective question clustering in the embodiment two of the present application;

[0019] Figure 5 is a flowchart of question deduplication in the embodiment two of the present application;

[0020] Figure 6 is a structural diagram of a question expansion device in Embodiment Three of the present application;

[0021] Figure 7 is a structural diagram of a server in Embodiment Four of the present application. DETAILED DESCRIPTION

[0022] The present application will be further described below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are intended to be merely illustrative of the present application and not limiting thereof. In addition, it should be noted that, for the sake of convenience, only the parts related to the present application are shown in the accompanying drawings rather than all the parts.

[0023] Embodiment One

[0024] Figure 1 is a flowchart of a question expansion method provided for Embodiment One of the present application. The present embodiment can be applied to expansion of questions in an interactive device. The method can be executed by a question expansion device and specifically includes the following steps:

[0025] S110, obtaining an interactive flow of a user and a device, the interactive flow including a question and an answer corresponding to the question.

[0026] With the rapid development of artificial intelligence, business question and answer systems and intelligent devices can replace human interaction with users to achieve the requirements of the users. The interactive flow refers to the content of the human-machine interaction conversation in the use process of the business question and answer system or the question and answer intelligent device, mainly composed of user input text questions (or converted text from user recorded voice, pictures, etc.) and answers returned by the system or robot according to the question content. The data has the characteristics of time sequence, short question content, discrete question content, etc. The interactive flow of the user and the intelligent device is obtained, and the questions in the preset question database are expanded according to the questions and answers in the interactive flow. Question expansion refers to supplementing the questions in the preset question database.

[0027] S120, determining an effective question in the question, grouping the effective question according to the answer corresponding to the effective question, and determining a candidate content set.

[0028] The questions raised by the user in the interactive log include valid questions and invalid questions, and the invalid questions are idle chat, a long string of special characters and numbers, short and meaningless words, and sensitive words. Therefore, the invalid questions in the interactive log need to be removed, and the valid questions are retained. Optionally, the valid questions in the questions are determined, including: determining the number of words in the question, and determining that the question is a valid question when the number of words is greater than a preset threshold. When the number of words in the question is less than the preset threshold, it indicates that the question is too short, and it is mostly idle chat or short and meaningless words. Therefore, the question with the number of words less than the preset threshold is determined as an invalid question and is removed, and the question with the number of words greater than the preset threshold is retained.

[0029] Optionally, the valid questions in the questions are determined, including: performing word segmentation processing on the questions to obtain word segmentation results; performing similarity calculation on the word segmentation results and words in a preset word library, removing questions corresponding to words with a similarity greater than a preset threshold, and obtaining valid questions; and the preset word library includes at least one of a preset sensitive word library, a preset meaningless word library, and a preset special character library. One or more of the preset word libraries is selected as needed to screen the questions in the interactive log. For example, sensitive words can be set as needed. Meaningless words, such as “ah” and “oh”, and special characters, such as “##” and “@@**”. The questions in the interactive log are processed by word segmentation to obtain question segmentation, and the question segmentation and the words in the preset word library are calculated for similarity. If the similarity exceeds the preset threshold, it indicates that the question is an invalid question and needs to be deleted. If the similarity is less than the preset threshold, it indicates that the question is a valid question and is temporarily retained.

[0030] Optionally, the valid questions are grouped according to answers corresponding to the valid questions, and a candidate content set is determined, including: sorting the questions in the interactive log in reverse order of time; determining whether the answer corresponding to the valid question is a valid answer in the order, and if so, establishing a candidate content set of the valid question, and if not, continuing to process the valid question in the next order; determining a first valid question with a valid answer in the current order and a second valid question with a valid answer in the next order, and adding a third valid question corresponding to an invalid answer between the first valid question and the second valid question to the candidate content set of the first valid question to group the valid questions. When the user interacts with the intelligent device or the business question and answer system, the question raised by the user may not be saved in the preset question database, so the intelligent device or the business question and answer system cannot give a valid answer. The question with a valid answer is generally followed by several invalid questions, but the candidate content set is constructed according to the question with a valid answer. Therefore, the valid question with a valid answer needs to be found first to construct the candidate content set, otherwise the question without a valid answer cannot be grouped.

[0031] The questions can be sorted in time in the interactive flow in reverse order, so as to find the questions with valid answers first. Whether the answer corresponding to the valid question is a valid answer is determined in turn according to the sorting order. Optionally, the valid answer is an answer in a preset answer database, and if the answer corresponding to the valid question is a valid answer, the valid question is given a preset identifier.

[0032] After determining that the valid question in the current order has a valid answer, the valid question is determined to be a first valid question. When a question in the next order with a valid answer is encountered, the valid question is determined to be a second valid question. After determining the question in the current order, whether the valid question in the next order has a valid answer is determined. If yes, the valid question in the next order is continuously processed. If no, the valid question is determined to be a third valid question and is added to the candidate content set of the first valid question. All the third valid questions corresponding to the invalid answers in order between the first valid question and the second valid question are added to the candidate content set of the first valid question, so as to group the valid questions. The above process is repeated until all the valid questions in the interactive flow are grouped.

[0033] Optionally, establishing the candidate content set of the valid question includes: determining whether the valid question in the current order with a valid answer is the same as the valid question in the candidate content set already established with a valid answer. If yes, the valid question in the next order with a valid answer is continuously determined. If no, the candidate content set of the valid question in the current order with a valid answer is established. When the answer corresponding to the question is determined to be a valid answer, the question is given a preset identifier, and whether the question is the same as the valid question in the candidate content set already established with a valid answer is determined according to the preset identifier. If yes, the corresponding candidate content set is established, and the question in the next order is continuously processed. If no, the candidate content set corresponding to the question is newly established, so as to group the valid questions according to the candidate content set.

[0034] S130, for each candidate content set, the valid questions in the candidate content set are clustered to obtain a clustering result set.

[0035] After the candidate content set is established, for each candidate content set, the valid questions therein are clustered according to semantics, and the valid questions with the same semantics are clustered together to obtain a clustering result. Optionally, for each candidate content set, the valid questions in the candidate content set are clustered to obtain a clustering result set, including: for each candidate content set, a main idea sentence is determined, and the valid questions are clustered according to the main idea sentence to obtain a clustering result. Optionally, the valid question in a preset order in each candidate content set can be determined as the main idea sentence, and the valid questions with the same semantics as the main idea sentence are clustered together. Optionally, the main idea sentence can be one or more.

[0036] Optionally, for each candidate content set, a main theme sentence is determined, valid questions are clustered according to the main theme sentence, and a clustering result is obtained, including:

[0037] For each candidate content set, a valid question in a first order is determined as a first main theme sentence, and valid questions in the candidate content set that have a similarity greater than a preset threshold with the first main theme sentence are clustered together with the first main theme sentence to obtain a clustering result. It is determined whether there are unclustered valid questions in the current candidate content set. If there are, a second main theme sentence is determined from the unclustered valid questions for clustering until the current candidate content set does not contain unclustered valid questions. A clustering result set is obtained. The valid question in the first order in each candidate content set can be determined as the first main theme sentence, and the similarity of the valid questions in the candidate content set with the first main theme sentence is sequentially determined in order. The similarity greater than the preset threshold indicates that the valid question has the same semantics as the first main theme sentence. The valid question is clustered together with the main theme sentence. When the similarity of the valid question is less than the preset threshold, it indicates that the semantics of the valid question is different from the first main theme sentence. It is determined whether the similarity of the valid question with other main theme sentences of the current candidate content set is greater than the preset threshold. If yes, clustering is performed. If no, the valid question is clustered as a new main theme sentence until the clustering of the valid questions in the current candidate content set is completed, and a clustering result set is obtained.

[0038] In S140, a target question set is determined according to the clustering result set, and the target question set is added to the preset question database to expand the questions in the preset question database.

[0039] The clustering result is further processed to obtain a target question set, and the target question set is added to the preset question database to expand the questions in the preset question database. Optionally, the target question set is determined according to the clustering result set, including: removing the repeated valid questions in the clustering result set to obtain a first question set; and determining the target question set according to the first question set. The clustering result set may contain repeated valid questions, which need to be removed to save the memory space of the preset question database. The first question set is obtained after removing the repeated valid questions. The target question set is determined according to the first question set.

[0040] Optionally, the target question set is determined according to the first question set, including: removing the questions that have appeared in the preset question database from the first question set to obtain a second question set; and determining the target question set according to the second question set. It is determined whether the valid question in the first question set exists in the preset question database. If yes, it does not need to be added to the preset question database. The questions that have not appeared in the preset question database are grouped into the second question set, and the target question set is obtained by further processing the second question set.

[0041] Optionally, the second question set is obtained by removing questions that have appeared in the preset question database from the first question set, including: determining a target number of valid questions in each clustering result in the first question set; when the target number exceeds a preset threshold, calculating the similarity between a theme sentence in the first question set and a question in the preset question database, and determining that the theme sentence and the valid questions clustered together with the theme sentence have appeared in the preset question database when the similarity exceeds a preset threshold; and removing the theme sentence and the valid questions clustered together with the theme sentence from the first question set to obtain the second question set. When the number of valid questions in each clustering result in the first question set exceeds the preset threshold, it indicates that the semantic question is frequently asked by users. Therefore, the similarity between the theme sentence in the clustering result whose number exceeds the preset threshold and a question in the preset question database is calculated. Since the valid questions clustered together have the same semantics, only the similarity between the theme sentence and the question in the preset question database is calculated. When the similarity exceeds the preset threshold, it is determined that the theme sentence and the valid questions clustered together with the theme sentence have appeared in the preset question database. When the similarity is less than the preset threshold, it is determined that the theme sentence and the valid questions clustered together with the theme sentence have not appeared in the preset question database. The valid questions that have appeared in the preset question database are removed from the first question set to obtain the second question set.

[0042] Optionally, the target number of valid questions in each clustering result in the first question set is determined, including: determining whether the first question set includes a valid question that is the same as the content of the theme sentence, and if yes, determining the number of theme sentences contained, and subtracting the number of valid questions that are the same as the content of the theme sentence from the number of valid questions in the first question set to obtain the target number; and if no, the number of valid questions in the first question set is the target number. When the first question set includes a valid question that is the same as the theme sentence, it indicates that the valid question is a repeated question. The number of valid questions in the first question set is subtracted by the number of valid questions that are the same in content to obtain the target number. If the first question set does not include a valid question that is the same as the content of the theme sentence, the number of valid questions in the first question set is determined as the target number.

[0043] Optionally, the target question set is determined according to the second question set, including: screening questions in the second question set according to the frequency of the questions in the second question set appearing in the clustering result set to obtain the target question set. The questions in the second question set are sorted according to the frequency of the questions in the second question set appearing in the clustering result set, and optionally, ascending or descending sorting is performed. Questions in a preset order range are determined as the target question set. For example, when descending sorting is performed, the top 50 questions are determined as the target question set from a total of 100 questions. The target question set is added to the preset question database to expand the questions.

[0044] The technical solution of this embodiment obtains the user-device interaction flow, which includes questions and corresponding answers; identifies valid questions; groups the valid questions according to their corresponding answers to determine a candidate content set; and removes invalid questions to improve the effectiveness of the questions and reduce data processing pressure. For each candidate content set, the valid questions in the candidate content set are clustered to obtain a clustering result set; questions with the same semantics are clustered together to facilitate expansion of questions with that semantics; a target question set is determined based on the clustering result set, and the target question set is added to a preset question database to expand the questions in the preset question database. This solves the problem of users not getting valid answers when interacting with interactive devices, improves the efficiency of collecting business question materials for customer service robots, effectively supplements the question database, reduces manual maintenance costs, and improves the question-answer matching rate.

[0045] Example 2

[0046] This embodiment is a preferred embodiment based on the above embodiments, and provides a further detailed explanation of the problem extension method.

[0047] During the Q&A interaction between customers and intelligent customer service robots, the content of user questions is highly uncertain, and a large number of invalid questions exist in the Q&A stream. Therefore, it is necessary to clean and filter the user question requests, retaining only the interaction stream with certain business meaning. For example... Figure 2 As shown, the specific steps include: obtaining the interaction flow, sequentially selecting the questions (i.e., request content Rqi) in the interaction flow, determining whether the request content Rqi is a meaningless question; if so, continuing to process the next sequential request content; otherwise, determining it as a valid question.

[0048] like Figure 3As shown, the session ID in the interactive stream is dimensioned, the session ID is set according to the user, indicating a complete session identification of a user. The valid questions in the interactive stream are sorted in descending order according to the event sequence, and the request content RQi (i=0, 1, 2,...) is obtained in turn according to the session ID, and the answer, i.e. the response content RPi (i=0, 1, 2,...), is obtained. It is judged whether the response content RPi is an invalid answer. If yes, the next (i+1) data is processed. If the response content RPi is a valid answer, the question ID corresponding to RPi is obtained. When the answer corresponding to the question is an effective answer in the preset answer database, the question ID corresponding to the question is assigned, and it is judged whether the candidate content set list in the candidate content set list exists. If yes, the next step is continued. If not, the candidate content set of the question ID is newly built. If the response content RP(i+1) exists, it is judged whether the response content RP(i+1) is a valid answer. If yes, the next (i+2) data is analyzed. If the response content RP(i+1) is an invalid answer, the request content RQ(i+1) is added to the candidate content set of the question ID corresponding to RP, until all questions are added to the corresponding candidate content set, and finally a candidate content set list dimensioned by the question ID is obtained.

[0049] As shown in Figure 4 , the candidate content list corresponding to the question ID is processed in a loop dimensioned by the question ID. The request content Ti in the content set is selected in turn, and the similarity between Ti and the main idea sentence Cj in the clustering result set corresponding to the current question ID is calculated, and it is judged whether the similarity between Ti and Cj is greater than a preset threshold. If the similarity is greater than the threshold, Ti is added to the similar question set corresponding to the main idea sentence Cj. For Ti and Cj in the second step, if the similarity is less than the preset threshold, it is further judged whether there is a clustering result corresponding to Cj+1. If not, a clustering result with Ti as the main idea sentence is newly built under the current question ID. For the clustering result of Cj+1 in the third step, the second step and the third step are continued to be processed until the clustering grouping processing of the request content Ti is completed, and a clustering result set list dimensioned by the question ID is obtained.

[0050] As shown in Figure 5 , the number of questions in the clustering result set is calculated. First, the number of questions in the clustering result set after deduplication is calculated, and it is judged whether the similar questions after deduplication exist and have the same content as the main idea sentence. If not, the number of questions in the similar question set is equal to the number of questions after deduplication. If yes, the number of questions in the similar question set is equal to the number of questions after deduplication of the clustering result set minus the number of questions with the same content as the main idea sentence.

[0051] The method for determining whether the clustering result is expired is defined as follows: according to the question ID in the clustering result, the corresponding question in the preset question database is obtained, and then the main sentence of the clustering result is compared with the preset question database in terms of similarity, and if there is a comparison result with high similarity, it indicates that the clustering result is expired.

[0052] The questions in the question ID corresponding clustering result set list are processed respectively according to the question ID dimension. The clustering results are selected from the clustering result set list in sequence, and it is judged whether the number of similar questions of the clustering result reaches a preset threshold value, if greater than the preset threshold value, it is continued to judge whether the clustering result is expired, and the clustering result set is retained if not expired. After the secondary cleaning and filtering of the clustering results corresponding to all question IDs are completed, the cleaned clustering result set list is finally output, the target question set is obtained, and the target question set is added to the preset question database to expand the questions in the database.

[0053] Embodiment three

[0054] Figure 6 A structure diagram of a question expansion device provided for the embodiment three of the application, the question expansion device comprising: an interactive flow acquisition module 310, a candidate content set determination module 320, a clustering result set determination module 330 and a question expansion module 340.

[0055] The interactive flow acquisition module 310 is configured to acquire the interactive flow of the current user and the smart device, and the interactive flow comprises questions and answers corresponding to the questions. The candidate content set determination module 320 is configured to determine the effective questions in the questions, group the effective questions according to the answers corresponding to the effective questions, and determine the candidate content set. The clustering result set determination module 330 is configured to cluster the effective questions for each candidate content set to obtain a clustering result set. The question expansion module 340 is configured to determine a target question set according to the clustering result set, and add the target question set to a preset question database to expand the questions in the database.

[0056] In the technical solution of the above embodiment, the candidate content set determination module 320 comprises:

[0057] The question word number determination unit is configured to determine the word number of the question, and determine the question as an effective question when the word number is greater than a preset threshold value.

[0058] In the technical solution of the above embodiment, the candidate content set determination module 320 comprises:

[0059] The word segmentation processing unit is configured to perform word segmentation processing on the question to obtain a word segmentation result.

[0060] The similarity calculation unit is configured to calculate the similarity between the word segmentation result and words in a preset word library, remove a question corresponding to a word with a similarity greater than a preset threshold, and obtain an effective question.

[0061] In the technical solution of the above embodiment, the candidate content set determination module 320 comprises:

[0062] The question sorting unit is configured to sort questions in a time reverse order according to the interaction flow.

[0063] The candidate content set establishment unit is configured to determine, in sequence according to the sorting order, whether an answer corresponding to an effective question is a valid answer, and if yes, establish a candidate content set of the effective question, and if no, continue to process the next order effective question.

[0064] The effective question grouping unit is configured to determine a first effective question with a valid answer in a current order and a second effective question with a valid answer in a next order, and add a third effective question with an invalid answer between the first effective question and the second effective question to the candidate content set of the first effective question, so as to group effective questions.

[0065] In the technical solution of the above embodiment, the candidate content set establishment unit comprises:

[0066] The effective question determination subunit is configured to determine whether a current effective question with a valid answer is the same as an effective question with a valid answer in an established candidate content set, and if yes, continue to determine a next order effective question with a valid answer, and if no, establish a candidate content set of the current effective question with a valid answer.

[0067] In the technical solution of the above embodiment, the clustering result set determination module 330 comprises:

[0068] The main idea sentence determination unit is configured to determine a main idea sentence for each candidate content set, cluster effective questions according to the main idea sentence, and obtain a clustering result.

[0069] In the technical solution of the above embodiment, the main idea sentence determination unit comprises:

[0070] The first main idea sentence determination subunit is configured to determine, for each candidate content set, a first order effective question as a first main idea sentence, and cluster, together with the first main idea sentence, an effective question in the candidate content set with a similarity greater than a preset threshold to the first main idea sentence, to obtain a clustering result.

[0071] The unclustered valid question determining subunit is configured to determine whether there is an unclustered valid question in the current candidate content set, and if so, determine a second main idea sentence from the unclustered valid question for clustering until the current candidate content set does not contain an unclustered valid question. A clustering result set is obtained.

[0072] In the technical solution of the above embodiment, the question expansion module 340 comprises:

[0073] The first question set determining unit is configured to remove the repeated valid questions in the clustered result set to obtain a first question set.

[0074] The target question set determining unit is configured to determine a target question set according to the first question set.

[0075] In the technical solution of the above embodiment, the target question set determining unit comprises:

[0076] The second question set determining subunit is configured to remove the questions that have appeared in the preset question database from the first question set to obtain a second question set.

[0077] The target question set determining subunit is configured to determine a target question set according to the second question set.

[0078] In the technical solution of the above embodiment, the second question set determining subunit comprises:

[0079] The target quantity determining subunit is configured to determine a target quantity of the valid questions of each clustered result in the first question set.

[0080] The similarity calculating subunit is configured to calculate the similarity between the main idea sentence in the first question set and the questions in the preset question database when the target quantity exceeds a preset threshold value, and determine that the main idea sentence and the valid questions clustered together with the main idea sentence have appeared in the preset question database when the similarity exceeds a preset threshold value.

[0081] The valid question removing subunit is configured to remove the main idea sentence and the valid questions clustered together with the main idea sentence from the first question set to obtain a second question set.

[0082] In the technical solution of the above embodiment, the target quantity determining subunit comprises:

[0083] The valid question determining subunit is configured to determine whether the first question set contains a valid question with the same content as the main idea sentence, and if so, determine the number of contained main idea sentences, and subtract the number of the valid questions with the same content as the main idea sentence from the number of the valid questions in the first question set to obtain a target quantity.

[0084] If not, the number of the valid questions in the first question set is the target quantity.

[0085] In the technical solutions of the above embodiments, the target problem set determination sub-unit includes:

[0086] The problem filtering subunit is used to filter the problems in the problem set according to the frequency of the occurrence of the problems in the second problem set in the clustering result set, so as to obtain the target problem set.

[0087] The technical solution of this invention obtains the user-device interaction flow, which includes questions and corresponding answers; identifies valid questions; groups the valid questions according to their corresponding answers to determine a candidate content set; and removes invalid questions to improve the effectiveness of the questions and reduce data processing pressure. For each candidate content set, the valid questions in the candidate content set are clustered to obtain a clustering result set; questions with the same semantics are clustered together to facilitate expansion of questions with that semantics; a target question set is determined based on the clustering result set, and the target question set is added to a preset question database to expand the questions in the preset question database. This solves the problem of users not receiving valid answers when interacting with interactive devices, improves the efficiency of collecting business question materials for customer service robots, effectively supplements the question database, reduces manual maintenance costs, and improves the question-answer matching rate.

[0088] The problem expansion device provided in the embodiments of the present invention can execute the problem expansion method provided in any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the method execution.

[0089] Example 4

[0090] Figure 7 This is a schematic diagram of the structure of a server provided in Embodiment 4 of the present invention, as shown below. Figure 7 As shown, the server includes a processor 410, a memory 420, an input device 430, and an output device 440; the number of processors 410 in the server can be one or more. Figure 7 Taking a processor 410 as an example; the processor 410, memory 420, input device 430, and output device 440 in the server can be connected via a bus or other means. Figure 7 Taking the example of a connection between China and Israel via a bus.

[0091] The memory 420, as a computer readable storage medium, can be used to store software programs, computer executable programs and modules, such as program instructions / modules corresponding to the problem expansion method in the embodiments of the present application (for example, the interactive flow acquisition module 310, the candidate content set determination module 320, the clustering result set determination module 330 and the problem expansion module 340 in the problem expansion device). The processor 410 executes various function applications and data processing of the server by running the software programs, instructions and modules stored in the memory 420, that is, implements the problem expansion method described above.

[0092] The memory 420 can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system and application programs required by at least one function; the data storage area can store data created according to the use of the terminal, etc. In addition, the memory 420 can include a high-speed random access memory, and can also include a non-volatile memory, such as at least one magnetic disk storage device, a flash memory device, or other non-volatile solid-state memory device. In some examples, the memory 420 can further include a memory remotely arranged with respect to the processor 410, which can be connected to the server through a network. Examples of the above network include but are not limited to the Internet, an intranet, a local area network, a mobile communication network and a combination thereof.

[0093] The input device 430 can be used to receive input digital or character information, and generate key signal input related to user settings and function control of the server. The output device 440 can include a display device such as a display screen.

[0094] Embodiment five

[0095] The embodiment five of the present application also provides a storage medium containing computer executable instructions, which when executed by a computer processor, are used to execute a problem expansion method, the method comprising:

[0096] Acquiring an interactive flow of a user with a device, the interactive flow including a question and an answer corresponding to the question;

[0097] Determining an effective question in the question, grouping the effective question according to an answer corresponding to the effective question, and determining a candidate content set;

[0098] For each candidate content set, clustering the effective questions in the candidate content set to obtain a clustering result set;

[0099] Determining a target question set according to the clustering result set, and adding the target question set to a preset question database to expand the questions in the preset question database.

[0100] Of course, the storage medium provided by the embodiment of the present application contains computer executable instructions, which are not limited to the method operations described above, but can also perform related operations in the problem expansion method provided by any embodiment of the present application.

[0101] Through the above description of the embodiments, those skilled in the art can clearly understand that the present application can be realized by software and necessary general hardware, and of course can also be realized by hardware, but in many cases the former is a better embodiment. Based on such understanding, the technical solutions of the present application can be embodied in the form of a software product, and the computer software product can be stored in a computer readable storage medium, such as a floppy disk, a read-only memory (ROM), a random access memory (RAM), a FLASH, a hard disk or an optical disk, etc., including a plurality of instructions to make a computer device (which can be a personal computer, a server, or a network device, etc.) execute the method described in each embodiment of the present application.

[0102] It is worth noting that the embodiments of the above problem expansion device include various units and modules only according to functional logic, but are not limited to the above division, as long as the corresponding functions can be realized; in addition, the specific names of each functional unit are only for easy mutual distinction, and do not limit the protection scope of the present application.

[0103] Note that the above is only the preferred embodiment of the present application and the technical principle applied. Those skilled in the art will understand that the present application is not limited to the specific embodiments described herein, and those skilled in the art can make various obvious changes, readjustments and substitutions without departing from the scope of the present application. Therefore, although the present application has been described in more detail through the above embodiments, the present application is not limited to the above embodiments, and can include more other equivalent embodiments without departing from the concept of the present application, and the scope of the present application is determined by the scope of the appended claims.

Claims

1. A method for extending a problem, characterized in that, include: The interaction flow between the user and the device is obtained. The interaction flow includes questions and answers corresponding to the questions. The questions include invalid questions, which are casual questions or questions with semantic information that are not related to business. Identify the valid questions among the questions, and group the valid questions according to the answers corresponding to the valid questions to determine the candidate content set; For each candidate content set, the valid questions in the candidate content set are clustered to obtain a clustering result set; The target question set is determined based on the clustering result set, and the target question set is added to the preset question database to expand the questions in the preset question database; The step of grouping the valid questions according to the answers to the valid questions to determine the candidate content set includes: Sort the questions in reverse chronological order according to the time in the interactive flow; The system sequentially determines whether the answers to valid questions are valid. If they are, a candidate content set for the valid questions is established, and a preset identifier is assigned to the valid questions. If not, the system continues to process the next sequence of valid questions. The valid answers are the answers in the preset answer database, and the valid answers are used to determine the preset identifier of the valid questions. Determine the first valid question with a valid answer in the current order, and the second valid question with a valid answer in the next order. Add the third valid question corresponding to the invalid answer in the order between the first valid question and the second valid question to the candidate content set of the first valid question, so as to group the valid questions. The establishment of the candidate content set for the valid question includes: Determine if the current valid question with a valid answer is the same as the valid questions with valid answers in the established candidate content set. If so, continue to determine the next valid questions with valid answers in the next order; otherwise, establish a candidate content set for the current valid question with a valid answer. Specifically, for each candidate content set, the valid questions in the candidate content set are clustered to obtain a clustering result set, including: For each candidate content set, determine the main idea sentence, and cluster the valid questions based on the main idea sentence to obtain the clustering results; Specifically, for each candidate content set, determining the main idea sentence, and clustering the valid questions based on the main idea sentence to obtain the clustering results includes: For each candidate content set, the first valid question in the first order is determined as the first topic sentence. The valid questions in the candidate content set whose similarity to the first topic sentence is greater than a preset threshold are clustered together with the first topic sentence to obtain the clustering result. Determine if there are any unclustered valid questions in the current candidate content set. If so, determine the second topic sentence from the unclustered valid questions and cluster them until the current candidate content set does not contain any unclustered valid questions, and obtain the clustering result set. The clustering results also include validity, and the method for judging validity is as follows: Obtain the question ID from the clustering results, and determine the corresponding question in the preset question database based on the question ID; compare the similarity between the topic sentence in the clustering results and the question in the preset question database, and determine the validity of the clustering results if the similarity is less than a preset threshold.

2. The method according to claim 1, characterized in that, The determination of valid questions among the questions includes: The number of characters in the question is determined. When the number of characters is greater than a preset threshold, the question is determined to be a valid question.

3. The method according to claim 1, characterized in that, The determination of valid questions among the questions includes: The problem is processed by word segmentation to obtain the word segmentation result; The word segmentation results are compared with the words in the preset dictionary for similarity calculation. Questions corresponding to words with similarity greater than a preset threshold are removed to obtain valid questions. The preset dictionary includes at least one of the following: a preset sensitive word library, a preset meaningless word library, and a preset special character library.

4. The method according to claim 1, characterized in that, The step of determining the target problem set based on the clustering result set includes: Remove duplicate valid questions from the clustering result set to obtain the first question set; The target question set is determined based on the first question set.

5. The method according to claim 4, characterized in that, Determining the target problem set based on the first problem set includes: A second set of questions is obtained by removing questions that have appeared in the preset question database from the first set of questions. The target problem set is determined based on the second problem set.

6. The method according to claim 5, characterized in that, The step of removing questions that have appeared in the preset question database from the first question set to obtain the second question set includes: Determine the target number of valid questions for each clustering result in the first problem set; When the number of targets exceeds a preset threshold, the similarity between the topic sentence in the first question set and the questions in the preset question database is calculated. When the similarity exceeds the preset threshold, it is determined that the topic sentence and the valid questions clustered with it appear in the preset question database. The topic sentence and the valid questions clustered with it are removed from the first question set to obtain the second question set.

7. The method according to claim 6, characterized in that, The determination of the target number of valid questions for each clustering result in the first problem set includes: Determine whether the first question set includes valid questions with the same content as the topic sentence. If so, subtract the number of valid questions with the same content as the topic sentence from the number of valid questions in the first question set to obtain the target number. If not, the number of valid questions in the first problem set is the target number.

8. The method according to claim 5, characterized in that, The step of determining the target problem set based on the second problem set includes: The questions in the second question set are filtered based on the frequency of their occurrence in the clustering result set to obtain the target question set.

9. A problem extension device, characterized in that, include: The interaction flow acquisition module is used to acquire the current user's interaction flow with the smart device, the interaction flow including questions and answers corresponding to the questions; The candidate content set determination module is used to determine the valid questions among the questions, group the valid questions according to the answers corresponding to the valid questions, and determine the candidate content set. The questions include invalid questions, which are casual questions or questions without business-related semantic information. The clustering result set determination module is used to cluster the valid questions for each candidate content set to obtain the clustering result set; The problem expansion module is used to determine the target problem set based on the clustering result set, and add the target problem set to the preset problem database to expand the problems in the database; The candidate content set determination module includes: The question sorting unit is used to sort questions in reverse chronological order according to the interactive flow. The candidate content set establishment unit is used to determine whether the answer corresponding to the valid question is a valid answer in the sorting order. If it is, a candidate content set for the valid question is established and a preset identifier is assigned to the valid question. If not, the next valid question in the next order is processed. The valid answer is an answer in the preset answer database and is used to determine the preset identifier of the valid question. The valid question grouping unit is used to determine the first valid question with a valid answer in the current order, and the second valid question with a valid answer in the next order, and to add the third valid question corresponding to the invalid answer in the order between the first valid question and the second valid question to the candidate content set of the first valid question, so as to group the valid questions. The candidate content set establishment unit includes: The valid question determination subunit is used to determine whether the currently valid question with a valid answer is the same as the valid questions with valid answers in the established candidate content set. If so, it continues to determine the next valid question with a valid answer in the next order; otherwise, it establishes a candidate content set for the currently valid question with a valid answer. The clustering result set determination module includes: The main idea sentence determination unit is used to determine the main idea sentence for each candidate content set, and to cluster the valid questions based on the main idea sentence to obtain the clustering results; The topic sentence determination unit includes: The first topic sentence determination sub-unit is used to determine the first topic sentence for each candidate content set, and to cluster the valid questions in the candidate content set that have a similarity greater than a preset threshold with the first topic sentence together with the first topic sentence to obtain the clustering result. The non-clustered valid question determination sub-unit is used to determine whether there are non-clustered valid questions in the current candidate content set. If so, the second topic sentence is determined from the non-clustered valid questions and clustered until the current candidate content set does not contain any non-clustered valid questions, and the clustering result set is obtained. The clustering results also include validity, and the method for judging validity is as follows: Obtain the question ID from the clustering results, and determine the corresponding question in the preset question database based on the question ID; compare the similarity between the topic sentence in the clustering results and the question in the preset question database, and determine the validity of the clustering results if the similarity is less than a preset threshold.

10. A server, characterized in that, The server includes: One or more processors; Storage device for storing one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the problem extension method as described in any one of claims 1-8.

11. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the problem extension method as described in any one of claims 1-8.

Images