Method and device for constructing customer service knowledge base

Abstract

This application discloses a method and apparatus for constructing a customer service knowledge base. The method includes: acquiring a detail image of a customer inquiring individual and identifying the text information carried in the detail image; determining whether the detail image is used to describe a target attribute based on the text information; if the detail image is used to describe the target attribute, generating a corresponding customer service question for the detail image based on the target attribute and the text information; generating a question-and-answer pair for the customer inquiring individual based on the customer service question and the detail image, and storing the question-and-answer pair in the customer service knowledge base.

Description

Technical Field

[0001] This specification relates to the field of Internet technology, and in particular to a method and apparatus for constructing a customer service knowledge base. Background Technology

[0002] With the rapid development of internet technology, intelligent customer service has been widely applied in various fields, establishing a fast and effective technical means for communication between enterprises and massive numbers of users based on natural language. To implement intelligent customer service, it is usually necessary to pre-build a customer service knowledge base, which stores the mapping relationship between questions and answers. For a user's question, the corresponding answer can be found in the customer service knowledge base and returned to the user. How to intelligently build an accurate customer service knowledge base has become a pressing technical problem to be solved. Summary of the Invention

[0003] In view of this, this specification provides a method and apparatus for constructing a customer service knowledge base.

[0004] Specifically, this specification is implemented through the following technical solution:

[0005] A method for constructing a customer service knowledge base includes:

[0006] Obtain a detailed image of the person seeking consultation and identify the text information carried in the detailed image;

[0007] Determine whether the detail image is used to describe the target attribute based on the text information;

[0008] When the detail image is used to describe the target attribute, a corresponding customer service question is generated for the detail image based on the target attribute and the text information.

[0009] Based on the customer service question and the detailed image, a question-and-answer pair is generated for the customer service recipient, and the question-and-answer pair is stored in the customer service knowledge base.

[0010] Optionally, determining whether the detail image is used to describe the target attribute based on the text information includes:

[0011] Determine whether the text information matches the target attribute;

[0012] Determine whether the text information matches the attribute value of the target attribute;

[0013] If the text information matches the target attribute and the attribute value, then the detail image is determined to be used to describe the target attribute.

[0014] Optionally, determining whether the text information matches the attribute value of the target attribute includes:

[0015] Obtain different attribute values ​​of the target attribute;

[0016] Determine whether the text information matches any of the different attribute values;

[0017] If the text information matches any attribute value, the attribute value of the target attribute is determined to match the text information.

[0018] Optionally, determining whether the detail image is used to describe the target attribute based on the text information includes:

[0019] If the text information does not match the target attribute, or does not match the attribute value, the target attribute is used as prior knowledge, and a machine reading comprehension MRC model is used to determine whether the detail image is used to describe the target attribute.

[0020] Optionally, generating corresponding customer service questions for the detail image based on the target attribute and the text information includes:

[0021] Based on the pre-generated mapping relationship between attributes and template questions, find the template question corresponding to the target attribute;

[0022] A question-and-answer matching model is used to determine whether the found template question matches the detail image;

[0023] If the template question matches the detail image, the matching template question is identified as the customer service question corresponding to the detail image.

[0024] Optionally, the template problem can be generated in the following ways:

[0025] The target attribute and its value are input into the question generation model, and the model outputs the customer service question corresponding to the target attribute, which serves as the template question for the target attribute.

[0026] A method for implementing intelligent customer service includes:

[0027] In response to user questions, the relevant consultation target for the user question is determined based on the user's browsing behavior;

[0028] Extract the attributes to be consulted from the user's question;

[0029] Search the customer service knowledge base for candidate questions corresponding to the consultation target and the consultation attribute;

[0030] Calculate the similarity between the user's question and the candidate questions, and if the similarity meets the similarity condition, return the detail image corresponding to the candidate question in the customer service knowledge base to the user;

[0031] The customer service knowledge base is constructed based on the aforementioned method.

[0032] An apparatus for building a customer service knowledge base, comprising:

[0033] The text recognition unit acquires a detailed image of the person seeking consultation and identifies the text information carried in the detailed image.

[0034] The attribute determination unit determines whether the detail image is used to describe the target attribute based on the text information.

[0035] The question generation unit generates a corresponding customer service question for the detail image based on the target attribute and the text information, when the detail image is used to describe the target attribute.

[0036] The knowledge base construction unit generates a question-and-answer pair for the customer service recipient based on the customer service question and the detail image, and stores the question-and-answer pair in the customer service knowledge base.

[0037] An intelligent customer service implementation device includes:

[0038] The object determination unit, in response to a user question, determines the target of the consultation based on the user's browsing behavior;

[0039] The attribute extraction unit extracts the attributes to be consulted from the user's question;

[0040] The candidate search unit searches the customer service knowledge base for candidate questions corresponding to the consultation object and the attribute to be consulted;

[0041] The customer service implementation unit calculates the similarity between the user's question and the candidate questions, and if the similarity meets the similarity conditions, returns the detail image corresponding to the candidate question in the customer service knowledge base to the user;

[0042] The customer service knowledge base is constructed based on the aforementioned method.

[0043] An apparatus for building a customer service knowledge base, comprising:

[0044] processor;

[0045] Memory used to store machine-executable instructions;

[0046] Specifically, the processor is prompted to: read and execute machine-executable instructions stored in the memory corresponding to the construction logic of the customer service knowledge base.

[0047] Obtain a detailed image of the person seeking consultation and identify the text information carried in the detailed image;

[0048] Determine whether the detail image is used to describe the target attribute based on the text information;

[0049] When the detail image is used to describe the target attribute, a corresponding customer service question is generated for the detail image based on the target attribute and the text information.

[0050] Based on the customer service question and the detailed image, a question-and-answer pair is generated for the customer service recipient, and the question-and-answer pair is stored in the customer service knowledge base.

[0051] An intelligent customer service implementation device includes:

[0052] processor;

[0053] Memory used to store machine-executable instructions;

[0054] Specifically, by reading and executing the machine-executable instructions stored in the memory corresponding to the implementation logic of the intelligent customer service system, the processor is prompted to:

[0055] In response to user questions, the relevant consultation target for the user question is determined based on the user's browsing behavior;

[0056] Extract the attributes to be consulted from the user's question;

[0057] Search the customer service knowledge base for candidate questions corresponding to the consultation target and the consultation attribute;

[0058] Calculate the similarity between the user's question and the candidate questions, and if the similarity meets the similarity condition, return the detail image corresponding to the candidate question in the customer service knowledge base to the user;

[0059] The customer service knowledge base is constructed based on the aforementioned method.

[0060] A computer-readable storage medium storing a computer program for causing a processor to execute the aforementioned method for constructing a customer service knowledge base.

[0061] By adopting the above implementation method, the process of building a customer service knowledge base eliminates the need for manual organization and uploading of the required information, significantly saving labor costs and improving the efficiency of knowledge base construction. Furthermore, this solution uses detailed images of the inquirer as answers to customer service questions. These images ensure high accuracy and eliminate the risk of information bias, fundamentally preventing inaccuracies and effectively guaranteeing the accuracy of the customer service knowledge base. Attached Figure Description

[0062] Figure 1 This is a flowchart illustrating an exemplary embodiment of a customer service knowledge base construction method.

[0063] Figure 2 This is a flowchart illustrating an exemplary embodiment of the present specification of a method for determining whether a detail diagram is used to describe a target attribute.

[0064] Figure 3 This is a flowchart illustrating another method for determining whether a detail diagram is used to describe a target attribute, as shown in an exemplary embodiment of this specification.

[0065] Figure 4 This is a flowchart illustrating an exemplary embodiment of a customer service question generation method.

[0066] Figure 5 This is a flowchart illustrating an exemplary embodiment of an intelligent customer service method described in this specification.

[0067] Figure 6 This is a hardware structure diagram of an electronic device containing a customer service knowledge base construction apparatus, as illustrated in an exemplary embodiment of this specification.

[0068] Figure 7 This is a block diagram illustrating an exemplary embodiment of a customer service knowledge base construction apparatus.

[0069] Figure 8 This is a hardware structure diagram of an electronic device containing an intelligent customer service implementation device, as illustrated in an exemplary embodiment of this specification.

[0070] Figure 9 This is a block diagram illustrating an exemplary embodiment of an intelligent customer service implementation device. Detailed Implementation

[0071] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numerals in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this specification. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this specification as detailed in the appended claims.

[0072] The terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to be limiting of this specification. The singular forms “a,” “the,” and “the” as used in this specification and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used herein refers to and includes any and all possible combinations of one or more of the associated listed items.

[0073] It should be understood that although the terms first, second, third, etc., may be used in this specification to describe various information, this information should not be limited to these terms. These terms are only used to distinguish information of the same type from one another. For example, without departing from the scope of this specification, first information may also be referred to as second information, and similarly, second information may also be referred to as first information. Depending on the context, the word "if" as used herein may be interpreted as "when," "when," or "in response to determination."

[0074] To implement intelligent customer service, it is typically necessary to pre-build a customer service knowledge base. This knowledge base stores the mapping relationship between questions and answers. For a user's question, the corresponding answer can be found in the knowledge base and returned to the user. For example, the similarity between a user's question and questions in the knowledge base can be calculated, and then the answer corresponding to the user's question can be found based on the similarity. Alternatively, a human customer service representative can retrieve the answer to the user's question from the knowledge base.

[0075] Regardless of the method used, building a customer service knowledge base is extremely important. One related technology involves users of intelligent customer service systems uploading relevant information, which is then aggregated to construct the knowledge base. For example, users collect and organize user feedback questions, then store the questions and answers in the knowledge base. However, this manual compilation method is labor-intensive, inefficient, and its accuracy cannot be guaranteed.

[0076] This manual provides a solution for building a customer service knowledge base, which can automatically generate question-and-answer pairs based on the details of the inquirer and store them in the customer service knowledge base. It has a high degree of intelligence and can also ensure accuracy.

[0077] Figure 1 This is a flowchart illustrating an exemplary embodiment of a customer service knowledge base construction method.

[0078] Please refer to Figure 1 The method for constructing the customer service knowledge base can be applied to a server or server cluster, and includes the following steps:

[0079] Step 102: Obtain the details image of the person seeking consultation and identify the text information carried in the details image.

[0080] In this manual, the objects of consultation may be goods or services sold on e-commerce platforms or laws and regulations published by government platforms. This manual does not impose any special restrictions on the specific categories of objects of consultation.

[0081] In this manual, e-commerce platforms typically provide detailed descriptions of their products or services, using either text or images. For example, for a specific product, configuration parameters can be displayed as text, while usage methods and internal structures can be shown through images. Similarly, government platforms can present the original text of laws and regulations using either images or text when interpreting them.

[0082] In this specification, the aforementioned images may be referred to as detail images of the inquiry target, and these detail images will serve as answers to customer service questions in the construction of the customer service knowledge base. When constructing the customer service knowledge base, all detail images of the inquiry target can be obtained first, and then the text information carried by each detail image can be identified.

[0083] Among them, OCR (Optical Character Recognition) technology can be used to identify the characters carried in the detail image, and then spatial clustering algorithm can be used to combine the identified characters to obtain semantically coherent text information (such as sentences).

[0084] The spatial clustering algorithm may be k-means, DBSCAN (Density-Based Spatial Clustering of Applications with Noise), OPTICS (Ordering points to identify the clustering structure), etc., and this specification does not impose any special restrictions on it.

[0085] In this manual, when using detailed images of the consulted object to build a customer service knowledge base, the detailed images can be filtered first.

[0086] In one example, because long images provide a poor reading experience for users, long images in the detail images can be filtered out. For instance, detail images longer than a length threshold can be filtered out, and long image filtering can be performed before character recognition.

[0087] In another example, after identifying characters in a detail image, detail images without characters can be filtered out, or detail images with fewer than a certain number of characters can be filtered out to filter out detail images that do not contain any substantial description. For example, if a detail image is a picture of a spokesperson using a product and does not contain any product description, it can be filtered out.

[0088] Of course, in other examples, the detail images can be filtered based on other dimensions, and this manual does not impose any special restrictions on this.

[0089] Step 104: Determine whether the detail image is used to describe the target attribute based on the text information.

[0090] This manual allows you to pre-set various attributes related to the clients being consulted, as well as the possible attribute values ​​for each attribute. For example, clients can be categorized according to their type, and attributes and attribute values ​​can be set for each category of clients.

[0091] For example, suppose the inquiry concerns products sold on an e-commerce platform, which can be categorized into clothing, digital products, baby products, and so on. Taking clothing as an example, users might focus on attributes such as material, style, color, size, and elasticity. For instance, material attributes could include multiple values ​​such as pure cotton, wool, silk, polyester, and nylon; style attributes for skirts could include multiple values ​​such as dress, skirt, and mini skirt.

[0092] In this specification, it can be determined whether a detail image is used to describe one or more pre-defined attributes (hereinafter referred to as target attributes) based on the text information in the detail image. For example, detail... Figure 1 Used to describe the material, details Figure 2 Used to describe style and material, etc.

[0093] In one example, it can be determined whether a detail image is used to describe a target attribute by checking whether the text information of the detail image includes the target attribute and any attribute value of the target attribute.

[0094] In another example, the MRC (Machine Reading Comprehension) model can also be used to determine whether a detail diagram is used to describe a target attribute.

[0095] If the detail diagram is used to describe one or more target attributes, step 106 can be performed.

[0096] Step 106: If the detail image is used to describe the target attribute, generate a corresponding customer service question for the detail image based on the target attribute and the text information.

[0097] Based on the judgment result of step 104 above, if the detail image is used to describe the target attribute, a corresponding customer service question can be generated for the detail image.

[0098] For example, a question generation model can be used to generate customer service questions corresponding to the detailed image.

[0099] For example, a mapping relationship between attributes and template questions can be pre-set, and then it can be determined whether the detail image matches a certain template question corresponding to the attribute it describes. If they match, the matching template question can be directly identified as the customer service question corresponding to the detail image.

[0100] Step 108: Generate a question-and-answer pair for the customer service recipient based on the customer service question and the detailed image, and store the question-and-answer pair in the customer service knowledge base.

[0101] In this manual, the mapping relationship between the consultation object identifier, detail image, and customer service question can be stored in the customer service knowledge base. To improve the subsequent query speed, the mapping relationship between the consultation object identifier, detail image, target attribute, and customer service question can also be stored. This manual does not impose any special restrictions on this.

[0102] As can be seen from the above description, this specification can determine whether a detail image is used to describe a target attribute based on the text information carried by the detail image. If the detail image is used to describe a target attribute, a corresponding customer service question can be generated for the detail image. Furthermore, a question-and-answer pair corresponding to the consultation can be generated based on the customer service question and the detail image and stored in the customer service knowledge base to realize the construction of the customer service knowledge base.

[0103] By adopting the above implementation method, the process of building a customer service knowledge base eliminates the need for manual organization and uploading of required information, significantly saving labor costs and improving the efficiency of knowledge base construction. Furthermore, this solution uses detailed images of the inquirer as answers to customer service questions. These images ensure high accuracy and eliminate the risk of information bias, fundamentally preventing inaccuracies and effectively guaranteeing the accuracy of the customer service knowledge base.

[0104] The following sections will detail the specific implementation process of this manual from two aspects: determining whether a detail image is used to describe a target attribute and generating customer service questions.

[0105] I. Determining whether a detailed image is used to describe the target attribute

[0106] In this specification, each pre-defined attribute can be used sequentially as a target attribute, and the appropriateness of the detail image for describing the target attribute can be determined accordingly. For example, all pre-defined attributes under the category to which the consulting object belongs can be obtained first, and then the determinations can be made sequentially.

[0107] Figure 2 This is a flowchart illustrating an exemplary embodiment of the present specification of a method for determining whether a detail diagram is used to describe a target attribute.

[0108] Please refer to Figure 2Taking a specific target attribute as an example, a method for determining whether a detail image of a consulting subject is used to describe the target attribute may include the following steps:

[0109] Step 202: Determine whether the text information carried by the detail image matches the target attribute.

[0110] In one example, when determining whether text information matches a target attribute, one can determine whether the target attribute exists in the text information.

[0111] Taking the target attribute as an example, we can determine whether the word "material" exists in the text information. For example, we can first segment the text information into words, and then make the judgment.

[0112] In another example, when determining whether text information matches a target attribute, it is also possible to determine whether there are synonyms for the target attribute in the text information.

[0113] Taking "material" as the target attribute, synonyms for "material" can include: material, usage, raw materials, etc. This allows us to determine if words like "material," "usage," and "raw materials" exist in the text. Alternatively, we can first segment the text and then make the judgment.

[0114] For example, it can be directly determined whether the aforementioned synonyms exist in the text information.

[0115] For example, text similarity can also be used to determine whether the semantic similarity between each word segmentation result of the text information and "material" is greater than a threshold.

[0116] Step 204: Determine whether the text information matches the attribute value of the target attribute.

[0117] In this step, several different attribute values ​​corresponding to the target attribute can be obtained, and then the text information can be judged in turn to determine whether it matches the attribute value.

[0118] Taking the target attribute as an example, its attribute values ​​may include: pure cotton, wool, silk, polyester, nylon, etc. In this step, it can be determined whether the text information matches these attribute values ​​in turn.

[0119] If the text information matches any attribute value of the target attribute, it can be determined that the text information matches the attribute value of the target attribute. During matching, if it has been determined that the text information matches a certain attribute value of the target attribute, subsequent matching of other attribute values ​​is not necessary, and it can be directly determined that the text information matches the attribute value of the target attribute.

[0120] If the text information does not match any attribute value of the target attribute, it can be determined that the text information does not match the attribute value of the target attribute.

[0121] The method for determining the attribute value matching process can be referred to the aforementioned attribute matching process, and will not be repeated here.

[0122] Step 206: If the text information matches the target attribute and the attribute value, determine that the detail image is used to describe the target attribute.

[0123] Based on the matching results of steps 202 and 204 above, if the text information matches the target attribute and also matches the attribute value, it is determined that the detail image is used to describe the target attribute, and a corresponding attribute tag can be added to the detail image.

[0124] If the text information does not match the target attribute, or if the text information does not match the attribute value of the target attribute, it can be determined that the detail image is not used to describe the target attribute.

[0125] In this embodiment, step 202 can be executed first. If it is determined that the text information does not match the target attribute, the matching of the attribute value of the target attribute in the subsequent step 204 can be skipped. Alternatively, steps 202 and 204 can be executed in parallel. Alternatively, step 204 can be executed first, and if it is determined in step 204 that the text information matches the attribute value of the target attribute, then step 202 can be executed. This embodiment does not impose any special restrictions on this.

[0126] This embodiment uses the method of whether the text information matches the target attribute and its attribute value to determine whether the detail image is used to describe the target attribute. It is simple to implement and relatively fast.

[0127] In this specification, except Figure 2 In addition to the judgment method shown, machine learning models can also be used to determine whether a detail image is used to describe a target attribute.

[0128] For example, the MRC model can be used to determine whether a detail image is used to describe a target attribute.

[0129] When using the MRC model to determine whether a detail image is used to describe a target attribute, the text information carried by the detail image and the target attribute can be input into the MRC model as input data. The target attribute can serve as prior knowledge for the MRC model to make its judgment.

[0130] Figure 3 This is a schematic diagram of the structure of an MRC model shown in an exemplary embodiment of this specification.

[0131] Please refer to Figure 3The MRC model may include an embedding layer 301, an encoding layer 302, an encoding layer 303, an attention layer 304, and a fully connected layer 305.

[0132] The embedding layer 301 is used to embed the text information and target attributes of the input detail image and convert them into corresponding vectors.

[0133] For example, for the text information, each character can be embedded to obtain a corresponding vector. For the target attribute, the target attribute can be embedded as a whole to obtain a vector corresponding to the target attribute.

[0134] The encoding layer 302 is used to encode the vector of text information output by the embedding layer 301.

[0135] The encoding layer 303 is used to encode the vector of target attributes output by the embedding layer 301.

[0136] The parameters of coding layer 302 and coding layer 303 are different.

[0137] Attention layer 304 is used to employ an attention mechanism to calculate an attention score by using the vector encoding of the target attribute and the vector encoding of each character in the text information, and then calculates a new vector for each character by using a weighted average method.

[0138] The formula for calculating the new vector is as follows:

[0139] score×token_emb_lstm+(1-score)×token-emb-out;

[0140] Here, score represents the attention score, token_emb_lstm represents the vector encoding obtained after encoding the text information vector, and token-emb-out represents an initialized vector.

[0141] Fully connected layer 305 is used to fully connect the new character vectors output by the attention layer and uses the Softmax function to output the probability value of whether the detail graph is used to describe the target attribute. For example, it outputs the probability value of each character in the text information representing the attribute value of the target attribute. When the probability value is greater than or equal to a threshold, it can be determined that the corresponding character represents the attribute value of the target attribute, that is, the detail graph is used to describe the target attribute. If the probability values ​​of all characters are less than the threshold, it can be determined that none of the characters can represent the attribute value of the target attribute, that is, the detail graph is not used to describe the target attribute.

[0142] When using the MRC model for judgment, the target attribute can be any of the preset attributes, meaning the MRC model is used to judge each preset attribute individually; or it can be a subset of all attributes, such as selecting some popular attributes and then using the MRC model for judgment. Selecting some popular attributes for judgment can effectively improve judgment efficiency and has a smaller impact on accuracy.

[0143] In this specification, it can be combined with Figure 2 and Figure 3 The methods shown can be used to jointly determine whether a detail image is used to describe a target attribute. For example, one could first use... Figure 2 The method shown is used for judgment. If the text information carried by the detail image does not match the target attribute, or the text information does not match the attribute value of the target attribute, then another method can be used. Figure 3 The MRC model shown is used for judgment. Figure 2 The judgment method shown is convenient and quick. Figure 3 The method shown combines semantic information for higher accuracy. This specification uses a combination of both to determine whether a detail image is used to describe a target attribute, achieving a balance between speed and accuracy.

[0144] II. Generating Customer Service Questions

[0145] In this specification, when it is determined that the detail image of the consultation object is used to describe the target attribute, a corresponding customer service question can be generated for the detail image based on the attribute values ​​and text information of the target attribute included in the detail image.

[0146] In one example, a question generation model can be used to generate corresponding customer service questions for the detail image. For instance, attribute values ​​and text information carried by the detail image can be input into the question generation model to obtain the customer service questions corresponding to the detail image.

[0147] Among them, the attribute values ​​can be found in the aforementioned Figure 2 The determination in step 204 of the illustrated embodiment can also be based on the foregoing. Figure 3 The output of the MRC model shown is determined.

[0148] The problem generation model can be the Transformer model, the classic seq2seq model (Sequence-to-sequence), etc.

[0149] In another example, because the problem generation model generates customer service questions relatively slowly, its timeliness may be poor when a large number of product detail images are deployed online. This example can pre-generate some template questions. When generating customer service questions for product detail images, the product detail images can be matched with various template questions, and the matched template questions can then be used as customer service questions, thereby improving the speed of customer service question generation.

[0150] In this example, a pre-set attribute and its value can be input into the question generation model to obtain customer service questions corresponding to the attribute. These customer service questions typically include various questions related to the attribute and may involve various attribute values. These customer service questions are used as template questions, and the mapping relationship between the attribute and these template questions is stored.

[0151] Taking the attribute "material" as an example, the material and its various attribute values ​​(e.g., pure cotton, wool, silk, polyester, nylon) can be input into the question generation model. For instance, each time a material and one attribute value are input, the model outputs several customer service questions, such as "Is it pure cotton?", "Is it wool?", and "What is the fabric made of?". These customer service questions can then be used as template questions corresponding to the material attribute and stored in association with the material attribute. Table 1 shows the mapping relationship between attributes and template questions.

[0152] property Template issues Material Template Question 1 - Template Question 10 shape Template Issues 11-25 … …

[0153] Table 1

[0154] Of course, Table 1 is only an example and may not be used in actual implementation.

[0155] Based on the mapping relationship between pre-generated attributes and template questions, the following can be adopted: Figure 4 The method shown is for generating customer service questions to create corresponding customer service questions for the product detail images.

[0156] Please refer to Figure 4 The generation of the customer service question may include the following steps:

[0157] Step 402: Based on the pre-generated mapping relationship between attributes and template questions, find the template question corresponding to the target attribute.

[0158] In this embodiment, if it has been confirmed that the detail image of the consultation object is used to describe the target attribute, the template question corresponding to the target attribute can be found based on the pre-generated mapping relationship between the attribute and the template question.

[0159] Taking the mapping relationship between attributes and template questions shown in Table 1 as an example, assuming that the target attribute is material, based on Table 1, the template questions corresponding to the material can be found to be template question 1-template question 10.

[0160] Step 404: Use a question-and-answer matching model to determine whether the found template question matches the detail image.

[0161] Step 406: If the found template question matches the detail image, the matched template question is determined as the customer service question corresponding to the detail image.

[0162] Based on the search results of step 402 above, it can be determined in turn whether each template question matches the detail image.

[0163] For example, the text information carried by the detail image and each template question can be input into the question-and-answer matching model. The model then outputs the matching degree between the detail image and the template question. If the matching degree is greater than a preset matching degree threshold, the detail image and the template question are considered a match, and the matched template question is identified as the customer service question for the detail image. If the matching degree is less than or equal to the preset matching degree threshold, the template question and the detail image are considered a mismatch, and a question generation model can be used to generate a corresponding customer service question for the detail image.

[0164] The question-answering matching model can be BERT (Bidirectional Encoder Representations from Transformer), DSSM (Deep Structured Semantic Models), or similar models.

[0165] It is worth noting that if multiple template questions match the details image, then all of these matching template questions can be identified as customer service questions corresponding to the details image.

[0166] For example, the text information carried in the details image includes "100% pure cotton fabric...", while the template questions corresponding to the material are "Is it pure cotton fabric?" and "What material is the fabric made of?". Both of these template questions match the details image, so template questions 1 and 2 can be identified as the customer service questions corresponding to the details image.

[0167] In this example, corresponding template questions are pre-generated for each attribute. When generating customer service questions for the detail images, the detail images are matched with each template question, and then the matched template questions are determined as the customer service questions corresponding to the detail images. Since the matching speed between questions and detail images is much faster than the speed at which the question generation model generates questions, when faced with a large number of detail images going online, using the template question matching method to generate customer service questions can greatly improve the efficiency and timeliness of customer service question generation.

[0168] This specification also provides a method for implementing intelligent customer service, which can be implemented based on the customer service knowledge base construction method described in the foregoing embodiments of this specification. Please refer to... Figure 5 The implementation method of the intelligent customer service may include the following steps:

[0169] Step 502: In response to the user's question, determine the target of the consultation based on the user's browsing behavior.

[0170] In this embodiment, when browsing products or services on an e-commerce platform, users can send their questions through the customer service portal. Similarly, when reading interpretations of laws and regulations on a government affairs platform, users can also send their questions through the customer service portal.

[0171] In response to the user's question, the relevant customer(s) can be identified. For example, on an e-commerce platform, the specific product the user wants to inquire about can be determined based on their browsing behavior.

[0172] For example, if the last item a user viewed before sending a user question was a dress with product ID 00075, that dress can be identified as the target of the user question.

[0173] For example, if a user triggers customer service through the customer service portal on the sales page of a dress with product ID 00075, then that dress can also be identified as the object of the user's inquiry.

[0174] After identifying the person to be consulted, their identifier can be obtained, such as their ID.

[0175] Step 504: Extract the attributes to be consulted from the user's question.

[0176] In this embodiment, the attribute to be consulted is the attribute that the user wants to consult. When extracting the attribute to be consulted, the user's question can be matched with preset attributes to determine the attribute to be consulted for the user's question. Of course, other methods can also be used to extract the attribute to be consulted, and this specification does not impose any special restrictions on this.

[0177] Step 506: Search the customer service knowledge base for candidate questions corresponding to the consultation object and the attribute to be consulted.

[0178] In this embodiment, the customer service knowledge base can store the mapping relationship between the consultation object identifier, detail image, target attribute, and customer service question.

[0179] The identifier for the consultation recipient can be the ID of the product or service, the ID of laws and regulations, etc.

[0180] Detailed images can be stored via their access links.

[0181] The target attribute is the attribute described in the detail image, and can be an attribute ID, etc.

[0182] Customer service questions can be the original text of the customer service question or a vector representation of the original text.

[0183] Consulting Client property Detailed image Customer Service Issues Item 00075 Material URL-1 What material Item 00075 Material URL-2 Is it pure cotton? Item 00075 shape URL-3 It's an A-line skirt. Item 00095 Material URL-4 Is it made of silk? … … … …

[0184] Table 2

[0185] Please refer to the mapping relationship diagram in the customer service knowledge base shown in Table 2. Assuming that the consultation object determined in step 502 is product 00075 and the consultation attribute extracted in step 504 is material, based on the customer service knowledge base shown in Table 2, two customer service questions can be extracted as candidate questions, namely "What material is it made of?" and "Is it pure cotton?".

[0186] Step 508: Calculate the similarity between the user's question and the candidate question, and if the similarity meets the similarity condition, return the detail image corresponding to the candidate question in the customer service knowledge base to the user.

[0187] Based on the extraction results of step 506, the similarity between the user question and each candidate question can be calculated. For example, the user question can be embedded and converted into a vector representation, and then the similarity between the vector representation of the user question and the vector representation of the customer service question can be calculated. For example, the Euclidean distance, cosine distance, etc., can be calculated to determine the similarity between the two.

[0188] In this embodiment, when the similarity condition is met, it can be determined that the user's question matches the customer service question, and then the detailed image corresponding to the matched customer service question can be returned to the user for viewing. For example, the detailed image can be returned based on the link corresponding to the customer service question stored in the customer service knowledge base.

[0189] The similarity conditions can be preset, such as similarity greater than a threshold.

[0190] As can be seen from the above description, this embodiment responds to user questions by searching the customer service knowledge base for candidate questions corresponding to the inquired object and the attribute to be inquired about. If the similarity between the user question and the candidate question meets the similarity condition, the detailed image corresponding to the candidate question is retrieved from the customer service knowledge base and returned to the user. Using this approach, returning the detailed image of the inquired object as the answer to the user results in higher accuracy and a better user experience.

[0191] Corresponding to the aforementioned embodiments of the customer service knowledge base construction method, this specification also provides embodiments of the customer service knowledge base construction apparatus.

[0192] The embodiments of the customer service knowledge base construction device described in this specification can be applied in electronic devices. The device embodiments can be implemented through software, hardware, or a combination of both. Taking software implementation as an example, as a logical device, it is formed by the processor of the electronic device loading the corresponding computer program instructions from non-volatile memory into memory for execution. From a hardware perspective, such as... Figure 6 The diagram shown is a hardware structure diagram of an electronic device containing the customer service knowledge base construction device of this manual, except... Figure 6 In addition to the processor, memory, network interface, and non-volatile memory shown, the electronic device in which the device is located in the embodiment may also include other hardware depending on the actual function of the electronic device, which will not be described in detail here.

[0193] Figure 7 This is a block diagram illustrating an exemplary embodiment of a customer service knowledge base construction apparatus.

[0194] Please refer to Figure 7 The customer service knowledge base construction device 700 can be applied to... Figure 6 The electronic device shown includes:

[0195] The text recognition unit 701 acquires a detailed image of the person seeking consultation and identifies the text information carried in the detailed image.

[0196] The attribute determination unit 702 determines whether the detail image is used to describe the target attribute based on the text information.

[0197] Question generation unit 703, when the detail image is used to describe the target attribute, generates a corresponding customer service question for the detail image based on the target attribute and the text information;

[0198] The knowledge base construction unit 704 generates a question-and-answer pair for the customer service recipient based on the customer service question and the detail image, and stores the question-and-answer pair in the customer service knowledge base.

[0199] Optionally, the attribute determination unit 702:

[0200] Determine whether the text information matches the target attribute;

[0201] Determine whether the text information matches the attribute value of the target attribute;

[0202] If the text information matches the target attribute and the attribute value, then the detail image is determined to be used to describe the target attribute.

[0203] Optionally, the attribute determination unit 702:

[0204] Obtain different attribute values ​​of the target attribute;

[0205] Determine whether the text information matches any of the different attribute values;

[0206] If the text information matches any attribute value, the attribute value of the target attribute is determined to match the text information.

[0207] Optionally, the attribute determination unit 702:

[0208] If the text information does not match the target attribute, or does not match the attribute value, the target attribute is used as prior knowledge, and a machine reading comprehension MRC model is used to determine whether the detail image is used to describe the target attribute.

[0209] Optionally, the problem generation unit 703:

[0210] Based on the pre-generated mapping relationship between attributes and template questions, find the template question corresponding to the target attribute;

[0211] A question-and-answer matching model is used to determine whether the found template question matches the detail image;

[0212] If the template question matches the detail image, the matching template question is identified as the customer service question corresponding to the detail image.

[0213] Optionally, the template problem can be generated in the following ways:

[0214] The target attribute and its value are input into the question generation model, and the model outputs the customer service question corresponding to the target attribute, which serves as the template question for the target attribute.

[0215] Corresponding to the aforementioned embodiments of the intelligent customer service implementation method, this specification also provides embodiments of the intelligent customer service implementation device.

[0216] The embodiments of the intelligent customer service implementation device described in this specification can be applied in electronic devices. The device embodiments can be implemented through software, hardware, or a combination of both. Taking software implementation as an example, as a logical device, it is formed by the processor of the electronic device loading the corresponding computer program instructions from non-volatile memory into memory for execution. From a hardware perspective, such as... Figure 8 The diagram shown is a hardware structure diagram of an electronic device containing the intelligent customer service implementation device described in this specification. (Except for...) Figure 8 In addition to the processor, memory, network interface, and non-volatile memory shown, the electronic device in which the device is located in the embodiment may also include other hardware depending on the actual function of the electronic device, which will not be described in detail here.

[0217] Figure 9 This is a block diagram illustrating an exemplary embodiment of an intelligent customer service implementation device.

[0218] Please refer to Figure 9 The intelligent customer service implementation device 900 can be applied in... Figure 8 The electronic device shown includes:

[0219] The object determination unit 901, in response to a user question, determines the consultation object to which the user question is directed based on the user's browsing behavior;

[0220] The attribute extraction unit 902 extracts the attributes to be consulted from the user's question;

[0221] The candidate search unit 903 searches the customer service knowledge base for candidate questions corresponding to the consultation object and the attribute to be consulted.

[0222] Customer service implementation unit 904 calculates the similarity between the user's question and the candidate question, and if the similarity meets the similarity condition, returns the detail image corresponding to the candidate question in the customer service knowledge base to the user; wherein, the customer service knowledge base is constructed based on the method described above in this specification.

[0223] The specific implementation process of the functions and roles of each unit in the above device can be found in the implementation process of the corresponding steps in the above method, and will not be repeated here.

[0224] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of the solution in this specification according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0225] The systems, devices, modules, or units described in the above embodiments can be implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer, which can take the form of a personal computer, laptop computer, cellular phone, camera phone, smartphone, personal digital assistant, media player, navigation device, email sending and receiving device, game console, tablet computer, wearable device, or any combination of these devices.

[0226] Corresponding to the aforementioned embodiments of the customer service knowledge base construction method, this specification also provides an apparatus for constructing a customer service knowledge base, the apparatus comprising: a processor and a memory for storing machine-executable instructions. The processor and memory are typically interconnected via an internal bus. In other possible implementations, the apparatus may also include an external interface to enable communication with other devices or components.

[0227] In this embodiment, the processor is prompted to: read and execute the machine-executable instructions stored in the memory corresponding to the construction logic of the customer service knowledge base.

[0228] Obtain a detailed image of the person seeking consultation and identify the text information carried in the detailed image;

[0229] Determine whether the detail image is used to describe the target attribute based on the text information;

[0230] When the detail image is used to describe the target attribute, a corresponding customer service question is generated for the detail image based on the target attribute and the text information.

[0231] Based on the customer service question and the detailed image, a question-and-answer pair is generated for the customer service recipient, and the question-and-answer pair is stored in the customer service knowledge base.

[0232] Optionally, determining whether the detail image is used to describe the target attribute based on the text information includes:

[0233] Determine whether the text information matches the target attribute;

[0234] Determine whether the text information matches the attribute value of the target attribute;

[0235] If the text information matches the target attribute and the attribute value, then the detail image is determined to be used to describe the target attribute.

[0236] Optionally, determining whether the text information matches the attribute value of the target attribute includes:

[0237] Obtain different attribute values ​​of the target attribute;

[0238] Determine whether the text information matches any of the different attribute values;

[0239] If the text information matches any attribute value, the attribute value of the target attribute is determined to match the text information.

[0240] Optionally, determining whether the detail image is used to describe the target attribute based on the text information includes:

[0241] If the text information does not match the target attribute, or does not match the attribute value, the target attribute is used as prior knowledge, and a machine reading comprehension MRC model is used to determine whether the detail image is used to describe the target attribute.

[0242] Optionally, generating corresponding customer service questions for the detail image based on the target attribute and the text information includes:

[0243] Based on the pre-generated mapping relationship between attributes and template questions, find the template question corresponding to the target attribute;

[0244] A question-and-answer matching model is used to determine whether the found template question matches the detail image;

[0245] If the template question matches the detail image, the matching template question is identified as the customer service question corresponding to the detail image.

[0246] Optionally, the template problem can be generated in the following ways:

[0247] The target attribute and its value are input into the question generation model, and the model outputs the customer service question corresponding to the target attribute, which serves as the template question for the target attribute.

[0248] Corresponding to the aforementioned embodiments of the intelligent customer service implementation method, this specification also provides an implementation apparatus for intelligent customer service, which includes: a processor and a memory for storing machine-executable instructions. The processor and memory are typically interconnected via an internal bus. In other possible implementations, the device may also include an external interface to enable communication with other devices or components.

[0249] In this embodiment, the processor is prompted to: read and execute the machine-executable instructions stored in the memory corresponding to the implementation logic of the intelligent customer service system.

[0250] In response to user questions, the relevant consultation target for the user question is determined based on the user's browsing behavior;

[0251] Extract the attributes to be consulted from the user's question;

[0252] Search the customer service knowledge base for candidate questions corresponding to the consultation target and the consultation attribute;

[0253] Calculate the similarity between the user's question and the candidate questions, and if the similarity meets the similarity condition, return the detail image corresponding to the candidate question in the customer service knowledge base to the user;

[0254] The customer service knowledge base is constructed based on the method described above in this specification.

[0255] Corresponding to the aforementioned embodiments of the customer service knowledge base construction method, this specification also provides a computer-readable storage medium storing a computer program that, when executed by a processor, performs the following steps:

[0256] Obtain a detailed image of the person seeking consultation and identify the text information carried in the detailed image;

[0257] Determine whether the detail image is used to describe the target attribute based on the text information;

[0258] When the detail image is used to describe the target attribute, a corresponding customer service question is generated for the detail image based on the target attribute and the text information.

[0259] Based on the customer service question and the detailed image, a question-and-answer pair is generated for the customer service recipient, and the question-and-answer pair is stored in the customer service knowledge base.

[0260] Optionally, determining whether the detail image is used to describe the target attribute based on the text information includes:

[0261] Determine whether the text information matches the target attribute;

[0262] Determine whether the text information matches the attribute value of the target attribute;

[0263] If the text information matches the target attribute and the attribute value, then the detail image is determined to be used to describe the target attribute.

[0264] Optionally, determining whether the text information matches the attribute value of the target attribute includes:

[0265] Obtain different attribute values ​​of the target attribute;

[0266] Determine whether the text information matches any of the different attribute values;

[0267] If the text information matches any attribute value, the attribute value of the target attribute is determined to match the text information.

[0268] Optionally, determining whether the detail image is used to describe the target attribute based on the text information includes:

[0269] If the text information does not match the target attribute, or does not match the attribute value, the target attribute is used as prior knowledge, and a machine reading comprehension MRC model is used to determine whether the detail image is used to describe the target attribute.

[0270] Optionally, generating corresponding customer service questions for the detail image based on the target attribute and the text information includes:

[0271] Based on the pre-generated mapping relationship between attributes and template questions, find the template question corresponding to the target attribute;

[0272] A question-and-answer matching model is used to determine whether the found template question matches the detail image;

[0273] If the template question matches the detail image, the matching template question is identified as the customer service question corresponding to the detail image.

[0274] Optionally, the template problem can be generated in the following ways:

[0275] The target attribute and its value are input into the question generation model, and the model outputs the customer service question corresponding to the target attribute, which serves as the template question for the target attribute.

[0276] Corresponding to the aforementioned embodiments of the intelligent customer service implementation method, this specification also provides a computer-readable storage medium storing a computer program that, when executed by a processor, performs the following steps:

[0277] In response to user questions, the relevant consultation target for the user question is determined based on the user's browsing behavior;

[0278] Extract the attributes to be consulted from the user's question;

[0279] Search the customer service knowledge base for candidate questions corresponding to the consultation target and the consultation attribute;

[0280] Calculate the similarity between the user's question and the candidate questions, and if the similarity meets the similarity condition, return the detail image corresponding to the candidate question in the customer service knowledge base to the user;

[0281] The customer service knowledge base is constructed based on the method described above in this specification.

[0282] The foregoing has described specific embodiments of this specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims may be performed in a different order than that shown in the embodiments and may still achieve the desired result. Furthermore, the processes depicted in the drawings do not necessarily require the specific or sequential order shown to achieve the desired result. In some embodiments, multitasking and parallel processing are possible or may be advantageous.

[0283] The above description is merely a preferred embodiment of this specification and is not intended to limit this specification. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this specification should be included within the scope of protection of this specification.

Claims

1. A method for constructing a customer service knowledge base, comprising: Obtain a detailed image of the person seeking consultation and identify the text information carried in the detailed image; Determine whether the detail image is used to describe the target attribute based on the text information; When the detail image is used to describe the target attribute, a corresponding customer service question is generated for the detail image based on the target attribute and the text information. Based on the customer service question and the detailed image, a question-and-answer pair is generated for the customer service recipient, and the question-and-answer pair is stored in the customer service knowledge base; The step of determining whether the detail image is used to describe the target attribute based on the text information includes: Determine whether the text information matches the target attribute; Determine whether the text information matches the attribute value of the target attribute; If the text information matches the target attribute and the attribute value, then the detail image is determined to be used to describe the target attribute.

2. The method according to claim 1, wherein determining whether the text information matches the attribute value of the target attribute includes: Obtain different attribute values ​​of the target attribute; Determine whether the text information matches any of the different attribute values; If the text information matches any attribute value, the attribute value of the target attribute is determined to match the text information.

3. The method according to claim 1, wherein determining whether the detail image is used to describe the target attribute based on the text information includes: If the text information does not match the target attribute, or does not match the attribute value, the target attribute is used as prior knowledge, and a machine reading comprehension MRC model is used to determine whether the detail image is used to describe the target attribute.

4. The method according to claim 1, wherein generating corresponding customer service questions for the detail image based on the target attribute and the text information includes: Based on the pre-generated mapping relationship between attributes and template questions, find the template question corresponding to the target attribute; A question-and-answer matching model is used to determine whether the found template question matches the detail image; If the template question matches the detail image, the matching template question is identified as the customer service question corresponding to the detail image.

5. The method according to claim 4, wherein the template problem is generated in a manner comprising: The target attribute and its value are input into the question generation model, and the model outputs the customer service question corresponding to the target attribute, which serves as the template question for the target attribute.

6. A method for implementing intelligent customer service, comprising: In response to user questions, the relevant consultation target for the user question is determined based on the user's browsing behavior; Extract the attributes to be consulted from the user's question; Search the customer service knowledge base for candidate questions corresponding to the consultation target and the consultation attribute; Calculate the similarity between the user's question and the candidate questions, and if the similarity meets the similarity condition, return the detail image corresponding to the candidate question in the customer service knowledge base to the user; The customer service knowledge base is constructed based on the method described in any one of claims 1-5.

7. A device for constructing a customer service knowledge base, comprising: The text recognition unit acquires a detailed image of the person seeking consultation and identifies the text information carried in the detailed image. The attribute determination unit determines whether the detail image is used to describe the target attribute based on the text information. The question generation unit generates a corresponding customer service question for the detail image based on the target attribute and the text information, when the detail image is used to describe the target attribute. The knowledge base construction unit generates a question-and-answer pair for the customer service recipient based on the customer service question and the detail image, and stores the question-and-answer pair in the customer service knowledge base; The attribute determination unit determines whether the detail image is used to describe the target attribute based on the text information, including: determining whether the text information matches the target attribute; and determining whether the text information matches the attribute value of the target attribute. If the text information matches the target attribute and the attribute value, then the detail image is determined to be used to describe the target attribute.

8. A device for implementing intelligent customer service, comprising: The object determination unit, in response to a user question, determines the target of the consultation based on the user's browsing behavior; The attribute extraction unit extracts the attributes to be consulted from the user's question; The candidate search unit searches the customer service knowledge base for candidate questions corresponding to the consultation object and the attribute to be consulted; The customer service implementation unit calculates the similarity between the user's question and the candidate questions, and if the similarity meets the similarity conditions, returns the detail image corresponding to the candidate question in the customer service knowledge base to the user; The customer service knowledge base is constructed based on the method described in any one of claims 1-5.

9. An apparatus for constructing a customer service knowledge base, comprising: processor; Memory used to store machine-executable instructions; Specifically, the processor is prompted to: read and execute machine-executable instructions stored in the memory corresponding to the construction logic of the customer service knowledge base. Obtain a detailed image of the person seeking consultation and identify the text information carried in the detailed image; Determine whether the detail image is used to describe the target attribute based on the text information; When the detail image is used to describe the target attribute, a corresponding customer service question is generated for the detail image based on the target attribute and the text information. Based on the customer service question and the detailed image, a question-and-answer pair is generated for the customer service recipient, and the question-and-answer pair is stored in the customer service knowledge base; The step of determining whether the detail image is used to describe the target attribute based on the text information includes: Determine whether the text information matches the target attribute; Determine whether the text information matches the attribute value of the target attribute; If the text information matches the target attribute and the attribute value, then the detail image is determined to be used to describe the target attribute.

10. A device for implementing intelligent customer service, comprising: processor; Memory used to store machine-executable instructions; Specifically, by reading and executing the machine-executable instructions stored in the memory corresponding to the implementation logic of the intelligent customer service system, the processor is prompted to: In response to user questions, the relevant consultation target for the user question is determined based on the user's browsing behavior; Extract the attributes to be consulted from the user's question; Search the customer service knowledge base for candidate questions corresponding to the consultation target and the consultation attribute; Calculate the similarity between the user's question and the candidate questions, and if the similarity meets the similarity condition, return the detail image corresponding to the candidate question in the customer service knowledge base to the user; The customer service knowledge base is constructed based on the method described in any one of claims 1-5.

11. A computer-readable storage medium storing a computer program for causing a processor to execute the method for constructing a customer service knowledge base as described in any one of claims 1-5.

Images