Customer service information generation method and system based on classification scene

Abstract

The invention discloses a customer service information generation method and system based on a classification scene. The method comprises the steps of obtaining current session data based on a preset memory window; performing vectorization operation on the current session data to obtain a target corpus; calling an AI model to analyze the target corpus, and classifying the current session data to preset scenes according to an analysis result, the preset scenes including a first scene, a second scene and a third scene, the first scene generating customer service information based on preset configuration information, the second scene generating customer service information based on a preset knowledge base, and the third scene generating customer service information based on a preset knowledge base; the third scene generates customer service information based on a preset information collection program; and based on the preset scene, generating customer service information corresponding to the current session data. The mode of generating the reply information through the scene type distinguishing model reduces the illusion influence of the model, greatly improves the accuracy of the customer service information, improves the model training efficiency, and reduces the model training cost.

Description

Technical Field

[0001] The present application relates to the field of computer technology, and in particular to a method for generating customer service information based on historical conversations, a method for generating customer service information based on classification scenarios, and a corresponding system. Background Art

[0002] With the development of artificial intelligence technology, automatic question-answering modules have gradually become an important part of the system to achieve information retrieval and human-computer interaction. In existing systems, automatic question-answering modules usually use keyword recognition and text similarity comparison to query answers to similar questions in historical conversations and then feed back the answers to users.

[0003] However, the processing method of the above-mentioned automatic question-answering module has the following problems: 1. Querying questions and answers based on keyword matching is difficult to apply in multilingual scenarios, and it requires a lot of manpower costs to label keywords in small languages; 2. Sentences with high or low text similarity may express very different meanings; 3. User questions queried based on keyword recognition and text similarity comparison have low question accuracy, and because they fail to combine the conversation context, they are prone to deviations in the aggregated answer information.

[0004] In addition, when the automatic question-answering module calls the generative large model to generate reply information, there is a possibility that the model may incorrectly recognize the user's intention, resulting in the generated reply information being inconsistent with the user's question and failing to meet the needs of the buyer; or, due to the hallucination problem of the model, there is a significant deviation between the reply information and the actual product information, misleading the buyer.

[0005] Therefore, how to improve the accuracy of customer service information generated by the system is a difficult problem that needs to be solved urgently in the e-commerce field.

[0006] Other technical issues related to this application will be further described later. The above content is only used to assist in understanding the technical solution of this application, and does not mean that all of the above content is prior art. Summary of the invention

[0007] The main purpose of the present invention is to provide a method and system for generating customer service information based on historical conversations, which can accurately judge user intent by combining historical conversation data and user preferences in multilingual scenarios, and automatically generate highly matched customer service information, thereby greatly improving the accuracy of customer service information. In addition, the present application also provides a method and system for generating customer service information based on classified scenarios, which can reduce the illusion effect of the model by performing scenario classification operations on user intent on the basis of accurately identifying user intent, thereby greatly improving the accuracy of customer service information.

[0008] To achieve the above objectives, the present application proposes a method and system for generating customer service information based on historical conversations, the method comprising: Step S1: Acquire historical session data, where the historical session data represents the content of the session between the buyer user and the customer service account; Step S2: performing a vectorization operation on the historical conversation data to obtain a first target corpus; Step S3: clustering the question sentences contained in the first target corpus according to vector similarity to obtain corpus clusters; Step S4: determining a corpus cluster whose cluster size reaches a preset size threshold as a target cluster, and determining a target question sentence corresponding to the target cluster; Step S5: determining the conversation context of the target question sentence in the first target corpus, and generating target answer information corresponding to the target question sentence based on the conversation context; Step S6: obtaining current customer service conversation data, and performing a vectorization operation on the customer service conversation data to obtain a second target corpus; Step S7: determining a target question sentence and target answer information corresponding to the second target corpus, and generating customer service information based on the target question sentence and the target answer information.

[0009] Other features and technical effects of this application are described in the following part of the specification. The technical problem solving ideas and related product design solutions of this application are: At present, the automatic question-answering module of the system usually uses keyword recognition and text similarity comparison to query similar questions and answers in historical conversations. However, the processing method of the automatic question-answering module has the following problems: 1. Keyword-matching query questions and answers is difficult to apply in multilingual scenarios, and requires a lot of manpower to annotate keywords in small languages. For example, cross-border e-commerce sellers handle consultation information from customers in multiple countries, and need to annotate keywords in each language. In addition, the person who performs the annotation operation needs to have a high level of proficiency in the language to ensure the accuracy of the annotation operation, so the labor cost requirement is high.

[0010] 2. Sentences with high or low text similarity may express very different meanings. For example, "I don't want to eat anymore" and "I can't eat" have high text similarity, but the meanings are very different; "I want to buy an anti-hair loss shampoo" and "Is there any anti-hair loss shampoo?" have low text similarity, but the meanings are very similar. Therefore, it is difficult to ensure the accuracy of the answer information when querying similar questions and answers based on text similarity.

[0011] 3. User questions queried based on keyword recognition and text similarity comparison have low accuracy, and similar questions and answers to the queries are not combined with the conversation context, which can easily lead to deviations in the final answer obtained by aggregating multiple question answers.

[0012] The applicant found that by performing vectorization operations on the system's conversation data based on the Embedding coding model, the conversation data of each language is converted into a vector representation of the same vector space (i.e., the target corpus), and then the target corpus is analyzed, thereby avoiding the need for cross-border e-commerce sellers to perform corpus annotation operations according to each language separately, reducing labor costs. In addition, since the original word order structure is broken after the conversation data is converted into the target corpus through vectorization operations, the semantic expression is not limited by the word order structure. Therefore, the system can accurately distinguish and identify sentences with high text similarity and very different meanings, or sentences with low text similarity and very similar meanings.

[0013] Furthermore, the applicant also found that the customer service information generated by the system has low accuracy, which is largely due to the inability to accurately judge the user's intention in the conversation scenario, and thus the difficulty in locating the user's question statements.

[0014] On this basis, the applicant proposed to use the community discovery algorithm to cluster the question statements contained in the target corpus, obtain multiple corpus clusters, and determine the target question statements corresponding to the corpus clusters with larger cluster sizes, which can greatly improve the AI ​​configuration efficiency of the system and reduce labor costs.

[0015] It is understandable that, since the corpus clusters correspond to question sentences, the corpus clusters with larger cluster sizes represent some question sentences with higher frequency of occurrence, and e-commerce sellers are required to provide corresponding reply information for these question sentences so that the system can call the AI ​​model to generate customer service information. In addition, for question sentences with lower frequency of occurrence, seller users can choose to postpone providing corresponding reply information, thereby applying human resources to questions that users are more concerned about.

[0016] Furthermore, the system obtains the current session data, performs vectorization operations on the current session data, and determines the target question statement corresponding to the current session data; queries the session context associated with the target question statement in the historical session data, and determines the reply information corresponding to the target question statement based on the session context; finally, the system generates customer service information based on the reply information, and displays the customer service information to the user, thereby accurately identifying the user's intention by combining the session context and the target question statement, and determining the corresponding target question statement and reply information according to the user's intention, so that the customer service information finally generated has extremely high accuracy.

[0017] In this way, the customer service information generation method based on historical conversations provided by the present application, when applied in multilingual scenarios, vectorizes the system's conversation data and clusters the question statements of the conversation data using a community discovery algorithm, thereby greatly improving the system's AI configuration efficiency, avoiding strong manual intervention, and reducing labor costs; and accurately judges user intentions by combining target question statements in the conversation context, automatically generating highly matching customer service information, thereby greatly improving the accuracy of customer service information.

[0018] Furthermore, the present application also provides a customer service information generation method based on classification scenarios, which is applied to the AI ​​module of the system, and the method includes: Step M1: obtaining current session data based on a preset memory window; Step M2: performing a vectorization operation on the current session data to obtain a target corpus; Step M3: calling the AI ​​model to analyze the target corpus, and classifying the current session data into preset scenarios according to the analysis results, wherein the preset scenarios include a first scenario, a second scenario, and a third scenario, wherein the first scenario generates customer service information based on preset configuration information, the second scenario generates customer service information based on a preset knowledge base, and the third scenario generates customer service information based on a preset information collection program; Step M4: Based on the preset scenario, generate customer service information corresponding to the current session data.

[0019] Other features and technical effects of this application are described in the following part of the specification. The technical problem solving ideas and related product design solutions of this application are: The existing intelligent customer service system of cross-border e-commerce uses intent recognition to generate customer service information. Usually, an AI model that supports multiple languages ​​is trained. This AI model recognizes user question information and generates reply information related to the user's question information.

[0020] However, the above method of generating customer service information has the following defects: 1. The model can easily misidentify the user's intent, resulting in the generated response information not matching the user's question.

[0021] 2. The model has an illusion problem. The response information generated based on the illusion speculation is obviously different from the actual product information, which misleads buyers and users.

[0022] 3. A general model needs to generate customer service information for user questions in various scenarios, so the training process requires a huge corpus, resulting in low model training efficiency and poor training results.

[0023] The applicant found that the model incorrectly identified user intent. The main reasons included the model's failure to analyze user questions in the context of the conversation data. Due to the omission of some information sent by the user, the model's recognition results of user intent were biased.

[0024] Then, since the AI ​​model is essentially an inference machine, when generating reply information, it tends to generate information that is highly correlated with the input information and has a higher probability. However, since the current e-commerce customer service system uses a general AI model to generate customer service information in response to user questions in a variety of scenarios, the information inferred by the AI ​​model based on user questions is too much and too complex. As a result, the AI ​​model may eventually generate reply information that does not match the actual product information due to hallucinations (i.e. excessive use of imagination).

[0025] In addition, since a general AI model needs to generate customer service information in response to user questions in a variety of scenarios, it is necessary to provide the general AI model with training corpora for a variety of scenarios; and in cross-border e-commerce business, it is necessary to provide corresponding training corpora in each language separately, resulting in low model training efficiency and poor training results.

[0026] On this basis, the applicant proposed to obtain multiple rounds of conversation data of the current conversation data based on a preset memory window, so that when the model identifies the user's intention when asking the user questions, it can combine the conversation context related to the user's questions, thereby improving the accuracy of identifying the user's intention.

[0027] Furthermore, the applicant proposed to classify user questions into scenarios and then adopt corresponding response strategies for different scenarios; and only call the model to generate reply information for some of the scenarios, and distinguish the way the model generates reply information according to the scenario type, which can effectively reduce the negative impact of the model's illusions and greatly improve the accuracy of customer service information generated by the model.

[0028] Specifically, the present application classifies user questions into the first scenario, the second scenario, and the third scenario. First, the first scenario does not involve complex information, and only requires responding to fixed terms or querying status information before responding. Therefore, the present application uses a classification model to classify user questions in the first scenario, and generates customer service information containing preset configuration information based on the classification results. Secondly, the second scenario involves complex information. The present application uses a generative large model to match user questions based on the second scenario with a preset knowledge base, thereby generating corresponding customer service information based on the preset knowledge base; wherein the preset knowledge base uses historical conversation data as training corpus to generate corresponding target question sentences and target reply information. Finally, the third scenario, as an information collection and feedback scenario, collects information provided by users based on a preset information collection program, including unboxing videos or user complaints, and then executes corresponding processing procedures (such as refunds and returns) or transfers to manual customer service.

[0029] It can be understood that in the above-mentioned processing scenarios for user questions, only the first and second scenarios need to call the model, and the first scenario only requires the model to provide pre-configuration information, and the difficulty of model training and model generation of customer service information is very low; while the second scenario requires the model to answer user questions based on a preset knowledge base, which is used to solve complex problems asked by users. It is more targeted and requires relatively less training corpus for the model.

[0030] In this way, the present application can, on the basis of accurately identifying user intentions, further perform scenario classification operations on user intentions, adopt corresponding response strategies for different scenarios, call the model to generate response information only for some of the scenarios, and distinguish the way the model generates response information according to the scenario type, thereby reducing the illusion effect of the model, greatly improving the accuracy of customer service information, and improving the efficiency of model training and reducing model training costs.

[0031] The present application also provides a server, which includes a memory and a processor. The system in the present application is stored in the memory, and the processor can run the operating instructions of each method step of the present application.

[0032] The present application also provides a computer device, which includes a memory and a processor. The system in the present application is stored in the memory, and the processor can execute the operating instructions of each method step in the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] The accompanying drawings are used to provide further understanding of the present application and do not constitute a limitation of the present application; the contents shown in the accompanying drawings may be real data of the embodiments and belong to the protection scope of the present application.

[0034] Figure 1 This is a schematic diagram of the process of generating customer service information by a customer service system after clustering question statements in one embodiment of the present application.

[0035] Figure 2 This is a flowchart of a method for generating customer service information based on historical conversations in one embodiment of the present application.

[0036] Figure 3 In one embodiment of the present application, a customer service system generates customer service information based on classified preset scenarios.

[0037] Figure 4 This is a flowchart of a method for generating customer service information based on classification scenarios in one embodiment of the present application. DETAILED DESCRIPTION

[0038] In order to make the purpose, technical solution and advantages of this application more clear, the following is a further detailed description of the embodiments of this application through specific implementation methods combined with the accompanying drawings. It should be understood that the specific embodiments described here are only used to explain this application and are not used to limit this application.

[0039] Figure 1 This is a schematic diagram of the process of generating customer service information by clustering question sentences in an e-commerce customer service system (hereinafter referred to as the system) in one embodiment of the present application. Figure 1 As shown, the AI ​​module of the system obtains the historical conversation data of the customer service module, and performs a vectorization operation on the historical conversation data to obtain the first target corpus. Then, the AI ​​module uses the community discovery algorithm to cluster the user question statements of the first target corpus to obtain multiple corpus clusters. Among them, the corpus cluster with a larger cluster size represents the user question statements with a higher frequency of occurrence. Next, the AI ​​module selects the corpus cluster with a larger cluster size, and determines the target question statement from the corpus cluster, that is, the target question statement has a higher frequency of occurrence and has a higher value. Then, the AI ​​module searches for the context conversation data corresponding to the target question statement (such as the customer service staff gave an answer after the buyer user consulted this type of question before), and generates the target reply information corresponding to the target question statement based on the context conversation data. When the system receives the buyer user's consultation information, the system performs a vectorization operation on the current conversation data to obtain the second target corpus; and the AI ​​module matches the second target corpus with the target question statement, and then determines the target reply information; finally, the AI ​​module generates the customer service information to reply to the buyer user based on the target reply information.

[0040] The e-commerce customer service system of the present application can be used as a subsystem of an e-commerce ERP system, an e-commerce platform system or other software systems, or it can be an independent customer service system.

[0041] Figure 2 A method for generating customer service information based on historical conversations in an embodiment of the present application is shown. The method mainly includes the following steps S1 to S7.

[0042] Step S1: Acquire historical session data, where the historical session data represents the content of the session between the buyer user and the customer service account.

[0043] Specifically, the system obtains historical conversation data of the customer service module, that is, the historical conversation data represents the conversation content between the buyer user and the customer service account, including the buyer user's question statements and the customer service staff's reply statements.

[0044] Step S2: performing a vectorization operation on the historical conversation data to obtain a first target corpus.

[0045] Specifically, the system uses an AI model to perform vectorization operations on historical conversation data, that is, converting conversation data in various languages ​​in the customer service module into vector representations in the same vector space, so that the semantic expression of historical conversation data is not limited by the word order structure.

[0046] Step S3: Clustering the question sentences included in the first target corpus according to vector similarity to obtain corpus clusters.

[0047] Specifically, the AI ​​module uses a community discovery algorithm to cluster user question sentences of the first target corpus according to vector similarity to obtain multiple corpus clusters; the multiple corpus clusters represent various types of inquiry sentences raised by buyer users.

[0048] Step S4: Determine the corpus cluster whose cluster size reaches a preset size threshold as the target cluster, and determine the target question sentence corresponding to the target cluster.

[0049] It can be understood that, since the corpus clusters correspond to question sentences, the corpus clusters with larger cluster sizes represent that some question sentences have a higher frequency of occurrence, that is, these user question sentences have higher value.

[0050] Therefore, in this embodiment, the target question sentence is determined based on the corpus cluster with a larger cluster size, and the corresponding target reply information is preferentially generated for the target question sentence, so that the system can call the AI ​​model to generate customer service information. For question sentences that appear less frequently, seller users can choose to postpone providing corresponding reply information, so as to apply AI configuration resources to questions that users are more concerned about.

[0051] Step S5: determining the conversation context of the target question sentence in the first target corpus, and generating target answer information corresponding to the target question sentence based on the conversation context.

[0052] Specifically, after determining the target question sentences with higher frequency, the system searches for the conversation context related to the target question sentences in the first target corpus, extracts the customer service staff's reply sentences in the conversation context, and then generates corresponding target reply information based on the reply sentences.

[0053] Step S6: Acquire current customer service conversation data, and perform vectorization operation on the customer service conversation data to obtain a second target corpus.

[0054] Specifically, when the system receives the buyer's user inquiry information, the system performs a vectorization operation on the current session data, so that the current session data is converted into a one-dimensional vector representation in the vector space where the first target corpus is located, that is, the current session data is converted into the second target corpus.

[0055] Step S7: determining a target question sentence and target answer information corresponding to the second target corpus, and generating customer service information based on the target question sentence and the target answer information.

[0056] Specifically, the system calls the AI ​​model to analyze the user intent of the second target corpus, determines the target question sentence and target reply information corresponding to the user intent, and then combines the target question sentence and target reply information to generate customer service information for replying to the buyer user.

[0057] As a feasible implementation method, the system performs word frequency analysis on the second target corpus, and focuses on displaying content related to functions or requirements that are mentioned many times by the buyer user in the generated customer service information. For example, if the buyer user mentions the camera function of the mobile phone many times in the conversation, the system will display a large amount of content related to the camera function of the mobile phone in the generated customer service information.

[0058] Further, in one embodiment, step S2 includes the following steps S2.1 and S2.2.

[0059] Step S2.1: Based on the historical session data, obtain commodity session data related to commodity information; Step S2.2: performing a vectorization operation on the commodity conversation data to obtain a first target corpus.

[0060] Specifically, the customer service conversation data of the system usually contains a large amount of meaningless content, including information content such as greetings, small talk, and emotional expressions sent by buyer users. In order to improve the quality of the training corpus for training the AI ​​module, improve the efficiency of AI model training, and improve the training effect, this embodiment performs a screening operation on the historical conversation data of the customer service module, only obtains the product conversation data related to the product information, and performs a vectorization operation on the product conversation data to obtain the first target corpus for configuring the AI ​​model.

[0061] Further, based on the above embodiment, step S2.2 includes the following step S2.21 and step S2.22.

[0062] Step S2.21: performing a filtering operation on the commodity session data based on the similarity between the session subject and the session data; Step S2.22: Perform a vectorization operation on the filtered commodity conversation data to obtain a first target corpus.

[0063] Specifically, in the customer service conversation data of the system, the customer service module will send an automatic reply to the buyer user when the manual customer service cannot handle it, the AI ​​model cannot recognize the answer, or when a specific sentence is received. It can be understood that the frequency of automatic replies is usually high and has no practical significance for AI configuration. This embodiment performs a filtering operation based on the similarity between the conversation subject and the conversation data, that is, highly repeated sentences issued by the customer service account are determined as automatic replies, and the relevant conversation data of the automatic replies are filtered. Finally, the commodity conversation data that has been filtered is vectorized, thereby obtaining a higher quality first target corpus.

[0064] Further, in one embodiment, step S5 includes the following steps S5.1 and S5.2.

[0065] Step S5.1: In the first target corpus, determining the conversation data whose time interval with the target question sentence is within a preset time range as the conversation context of the target question sentence; Step S5.2: Generate target answer information corresponding to the target question sentence based on the conversation context.

[0066] Specifically, the system searches for conversation data within a preset time range and corresponding to the target question statement in the first target corpus. For example, the system searches for conversation data within a preset time range from the target question statement, and then determines the customer service response statement corresponding to the target question statement in the conversation data, and aggregates the customer service response statement to generate target reply information.

[0067] In this way, this embodiment stipulates that when the system configures target response information for a target question statement, the target response information needs to be generated based on the context within the historical conversation scenario, so as to ensure the integrity and accuracy of the target response information to the greatest extent.

[0068] Further, based on the above embodiment, step S5.2 includes the following steps S5.21 to S5.23.

[0069] Step S5.21: determining corresponding candidate answer information in a plurality of conversation scenarios in the first target corpus based on the conversation context; Step S5.22: If it is detected that there are multiple candidate answer information, the multiple candidate answer information is pushed to the user confirmation interface; Step S5.23: In response to a first confirmation instruction from the user confirmation interface, determine that candidate reply information corresponding to the first confirmation instruction is the target reply information.

[0070] In this embodiment, the system finds multiple conversation scenarios corresponding to the target question sentence in the first target corpus, and the conversation scenarios contain conversation contexts related to the target question sentence; and the system aggregates multiple candidate answer information based on the multiple conversation scenarios, and the multiple candidate answer information are not completely consistent, and even express mutually exclusive and opposing meanings. On this basis, the system pushes the multiple candidate answer information to the user confirmation interface, and the user confirmation interface is operated and managed by the buyer user. When the buyer user interacts with the user confirmation interface and triggers the first confirmation instruction to select the corresponding candidate answer information, the system determines the candidate answer information corresponding to the first confirmation instruction as the target answer information.

[0071] Thus, in this embodiment, if the system aggregates multiple candidate reply information, the decision-making power for the candidate reply information is given to the buyer user, so that the customer service information generated by the system meets the original intention of the buyer user to the greatest extent.

[0072] Further, based on the above embodiment, after step S5.21, when it is detected that there are multiple candidate reply information, the method also includes the following steps S5.24 and S5.25.

[0073] Step S5.24: Detecting the occurrence frequencies of the plurality of candidate answer information, and performing a sorting operation on the plurality of candidate answer information based on the occurrence frequencies; Step S5.25: Determine target reply information based on the sorting result.

[0074] As a feasible implementation method, when the system aggregates multiple candidate reply information, and the buyer user cannot select and confirm the multiple candidate reply information immediately, the system will sort the occurrence frequencies of the multiple candidate reply information and determine the target reply information with the candidate reply information with higher occurrence frequency. For example, if the system aggregates reply information A, reply information B, and reply information C, among which reply information B appears the most times in the historical session data, the system determines that the seller's customer service staff tends to approve reply information B, and determines reply information B as the target reply information.

[0075] Furthermore, in one embodiment, after step S7, the method further includes the following steps S7.1 and S7.2.

[0076] Step S7.1: Push the customer service information to the user confirmation interface; Step S7.2: In response to a second confirmation instruction from the user confirmation interface, the customer service information is sent to the buyer user.

[0077] Specifically, in this embodiment, after the system automatically generates customer service information corresponding to the current session data, it pushes the customer service information to the user confirmation interface operated and managed by the seller user, and sends it to the buyer user only after the seller user confirms it. In this way, this embodiment gives the seller user the right to decide on the reply to the buyer user, so that the reply to the buyer user's customer service information can respect the seller user's wishes to the greatest extent.

[0078] As a feasible implementation method, when the system detects that there is a preset authorization agreement between the system and the seller's user account, the system will send the automatically generated customer service information directly to the buyer user without the need for confirmation by the seller user, thereby improving the efficiency of customer service information response.

[0079] Furthermore, in one embodiment, after step S6, the method further includes the following steps S7.3 to S7.5.

[0080] Step S7.3: when the target question sentence and target answer information corresponding to the second target corpus cannot be detected, generating a preset question sentence corresponding to the second target corpus; Step S7.4: marking the preset question statement; and, Step S7.5: Send a preset reply statement to the buyer user, or transfer the current customer conversation to a manual reply.

[0081] In this embodiment, the system cannot find the target question sentence and target reply information that match the second target corpus, that is, the user question sentences contained in the second target corpus are relatively niche, and the AI ​​module does not configure the corresponding reply information for the user question sentence; at this time, the system analyzes the second target corpus, generates the corresponding preset question sentence, and performs the marking operation. In the subsequent operation process, if the customer service staff replies to questions similar to the preset question sentence with the mark, the system will establish an association relationship between the reply content provided by the customer service staff and the preset question sentence, and use it as training corpus to further train the AI ​​model of the customer service module.

[0082] In this way, this embodiment records user questions that the system's AI model cannot answer immediately in the current form, and after the customer service staff subsequently answer similar questions, the user questions and the customer service staff's replies are used as new training materials to train the AI ​​model, so that the AI ​​model is continuously iterated and optimized.

[0083] Further, in one embodiment, step S7 includes the following steps S7.6 to S7.8.

[0084] Step S7.6: Detecting the number of characters in the user's sentence in the second target corpus, and determining a memory window based on the number of characters, wherein the memory window includes multiple rounds of conversations of the second target corpus, and the number of characters in the user's sentence is negatively correlated with the size of the memory window; Step S7.7: analyzing the conversation data contained in the memory window based on a preset attenuation factor, wherein the preset attenuation factor represents a weight change rule corresponding to each round of conversation in the memory window; Step S7.8: Determine the target question sentence and target answer information corresponding to the memory window according to the analysis result, and generate customer service information based on the target question sentence and the target answer information.

[0085] The applicant found that when the customer service module of the system receives information sent by buyer users, due to the different communication habits of each buyer user, the system uses a fixed memory window combined with the conversation context, which cannot fully and accurately identify the user's intention.

[0086] For example, the communication habits of buyer user A are: "I want a mobile phone with top-level chip performance and good camera performance" (first line of information), "the budget for the mobile phone is less than 6,000 yuan" (second line of information).

[0087] The communication habits of buyer B are: "I want a mobile phone" (first line of information), "I want a top-level chip performance" (second line of information), "good camera performance" (third line of information), "mobile phone budget" (fourth line of information), and "less than 6,000 yuan" (fifth line of information).

[0088] It can be seen that the buyer user A mentioned above is accustomed to sending the demand after editing a complete sentence, while the buyer user B is accustomed to sending the demand separately after distributing it in multiple sentences; if the customer service module uses a fixed-size memory window combined with the conversation context, such as the memory window contains three rounds of context conversations, the customer service module will miss some information when identifying the intention of the buyer user B (the customer service module will not recognize information beyond the scope of three rounds of conversations), resulting in deviations in the customer service information generated to reply to the buyer user B. In addition, if the customer service module uses a larger memory window, such as the memory window contains thirty rounds of context conversations, it will cause the system to process a large amount of data, which will impose a heavy burden on the system operation, and there will be a significant delay in the buyer user's reply to the customer service information, which will seriously affect the user experience.

[0089] On this basis, the applicant proposed to adjust the memory window according to the number of characters in the user's sentence, so as to ensure that the customer service module can fully and accurately identify the buyer's user intentions without placing too much burden on the system.

[0090] For example, the customer service module uses an initial memory window that includes three rounds of context conversations. However, when the customer service module detects that the number of characters in a single sentence of buyer user B is less than 10 characters, the customer service module automatically adjusts the memory window to include five rounds of context conversations, or eight rounds of context conversations, so that the system can fully and accurately identify the intention of buyer user B without processing too much data.

[0091] Furthermore, the applicant found that when the customer service module uses a larger memory window, since the memory window contains more rounds of dialogue, it is easy for the AI ​​model to find it difficult to grasp the key points of the conversation, resulting in deviations in the AI ​​model's recognition of user intent. Therefore, the applicant proposed that the customer service module analyze the conversation data of the memory window based on a preset attenuation factor, that is, assign different weights to each round of dialogue contained in the memory window according to the preset attenuation factor, so that the AI ​​model's recognition results of user intent can be closer to current user needs.

[0092] For example, assuming the preset attenuation factor is 0.8, the weight of the current round of dialogue information is 1, the weight of the previous round is 0.8, the weight of the previous round is 0.64, and so on. When the AI ​​model processes user intent and generates answers, it performs weighted processing based on the weight of the information, so that recent dialogue information has a greater impact on the current decision of the AI ​​model.

[0093] In this way, the customer service module of this embodiment adjusts the size of the memory window according to the characters of the user's sentence, and when the memory window contains a large number of rounds of dialogue, it assigns weights to each round of dialogue based on a preset attenuation factor, thereby ensuring that the AI ​​model can fully obtain the question information sent by the user and accurately analyze the user's intentions, so that the customer service information finally generated has higher accuracy.

[0094] As a feasible implementation method, the customer service module marks the conversation content contained in the above-mentioned memory window with conversation information such as current user intention, consultation type, and user feedback, and displays the conversation information in the background visualization interface so that the customer service personnel of the buyer user can understand the conversation status in real time and optimize and upgrade the AI ​​model.

[0095] As a feasible implementation, the customer service module is configured with anomaly detection rules. For example, when the buyer's question cannot be accurately understood by the system for a long time (the confidence level of intent recognition is lower than a certain threshold) or when there are too many rounds of dialogue but no effective results are achieved (such as failure to successfully recommend products or solve problems, resulting in repeated inquiries by users), an abnormal warning is triggered. For example, if the system's confidence level of user intent recognition is lower than 30% after 5 consecutive rounds of dialogue, an early warning notification is sent to the customer service staff, indicating that manual intervention may be required.

[0096] In addition, the customer service module monitors changes in user emotions. When it detects that the user is emotionally excited or dissatisfied (for example, words or sentences expressing dissatisfaction appear), the current conversation is marked and the dialogue strategy is adjusted in a timely manner to give priority to user emotional issues: such as adjusting the dialogue style of the customer service module and providing solutions, or arranging for manual customer service to respond first.

[0097] In one embodiment, customer service information is generated based on a system call to an AI model. The AI ​​model has an illusion problem, which causes a significant deviation between the reply information and the actual product information. The present application also provides a method for generating customer service information based on a classification scenario.

[0098] Figure 3 FIG. 1 shows a schematic diagram of a process in which the system generates customer service information based on classified preset scenarios in one embodiment of the present application. Figure 3 As shown, the AI ​​module of the system obtains the current session data of the customer service module based on a memory window of a preset size, and performs vectorization operations on the current session data to obtain the target corpus. Then, the AI ​​module recognizes the user intent of the target corpus, and classifies the current session into the first scenario, the second scenario, and the third scenario according to the recognition results. Among them, the first scenario does not involve complex information, and the target reply information is determined based on the preset configuration information; the second scenario involves complex information, and a generative large model is used to determine the target reply information based on the preset knowledge base; the third scenario is an information collection and feedback scenario, and the target reply information is determined based on the preset information collection program. Finally, the system generates customer service information containing the above-mentioned target reply information, and sends the customer service information to the user.

[0099] Figure 4 A customer service information generation method based on a classification scenario in an embodiment of the present application is shown, and the method mainly includes the following steps M1 to M7.

[0100] Step M1: Acquire current session data based on a preset memory window.

[0101] Specifically, the preset memory window is used to limit the scope of the AI ​​model's understanding of the conversation data. The system obtains the current conversation data of the customer service module based on the memory window of the preset size. For example, the system obtains the conversation content of the previous and next three rounds of the current conversation of the customer service module.

[0102] Step M2: Perform a vectorization operation on the current session data to obtain a target corpus.

[0103] Specifically, the system uses an AI model to perform vectorization operations on the current conversation data, that is, converting the conversation data in each language in the customer service module into vector representations in the same vector space, so that the semantic expression of the current conversation data is not limited by the word order structure.

[0104] Step M3: Call the AI ​​model to analyze the target corpus, and classify the current conversation data into preset scenarios based on the analysis results. The preset scenarios include a first scenario, a second scenario, and a third scenario. The first scenario generates customer service information based on preset configuration information, the second scenario generates customer service information based on a preset knowledge base, and the third scenario generates customer service information based on a preset information collection program.

[0105] Specifically, the present application classifies user questions into the first scenario, the second scenario, and the third scenario. First, the first scenario does not involve complex information, and only requires responding to fixed terms or querying status information before responding. Therefore, the present application uses a classification model to classify user questions in the first scenario, and generates customer service information containing preset configuration information based on the classification results. Secondly, the second scenario involves complex information. The present application uses a generative large model to match user questions based on the second scenario with a preset knowledge base, thereby generating corresponding customer service information based on the preset knowledge base; wherein the preset knowledge base uses historical conversation data as training corpus to generate corresponding target question sentences and target reply information. Finally, the third scenario, as an information collection and feedback scenario, collects information provided by users based on a preset information collection program, including unboxing videos or user complaints, and then executes corresponding processing procedures (such as refunds and returns) or transfers to manual customer service.

[0106] Step M4: Based on the preset scenario, generate customer service information corresponding to the current session data.

[0107] Specifically, based on the preset scenario corresponding to the target corpus, the system adopts a corresponding processing method to generate customer service information containing target reply information, and sends the customer service information to the user.

[0108] Furthermore, in one embodiment, the first scenario includes chat interaction, activity information, process consultation, and logistics inquiry.

[0109] Specifically, when a buyer sends a greeting message, information unrelated to the product or order, inquires about promotional information, order status, or logistics status in the customer service window, there is no need to train and call a specific generative large model to process it, as it does not involve complex information processing. Instead, you only need to call the classification model to reply to the user based on the pre-configured information.

[0110] Furthermore, in one embodiment, the second scenario includes product consultation.

[0111] Specifically, when a buyer asks questions about product-related content in the customer service window, including product recommendations, product parameters, product usage tutorials, etc., due to the complex information processing involved, the system calls the generative big model to identify the question statements contained in the current session data, and retrieves the target reply information corresponding to the current session data in the preset knowledge base, and replies to the user based on the target reply information.

[0112] Further, in one embodiment, the third scenario includes after-sales processing.

[0113] Specifically, when buyers raise after-sales demands such as returns, exchanges, refunds, or complaints at the customer service window, there is no need to train and call specific generative large models for processing, as it does not involve complex information processing and cannot meet user needs at the information communication level. On this basis, the system starts the preset information collection program, and asks users to "send the type of product problem", "send the order number", or "send the unboxing video", and executes the corresponding after-sales processing procedures based on user feedback.

[0114] Furthermore, in one embodiment, the method further includes the following steps M3.1 to M3.4.

[0115] Step M3.1: Cluster the question sentences in the historical conversation data to obtain corpus clusters; Step M3.2: Determine the corpus cluster whose cluster size reaches a preset size threshold as the target cluster, and determine the target question sentence corresponding to the target cluster; Step M3.3: Determine the target answer information corresponding to the target question sentence in the historical conversation data; Step M3.4: Generate a preset knowledge base based on the target question sentence and the target answer information.

[0116] Specifically, in this embodiment, after clustering the question sentences based on the historical conversation data, the system selects the question sentences with higher frequency and are representative as the target question sentences, and determines the target answer information corresponding to the target question sentences in the historical conversation data, and then constructs a preset knowledge base based on the target question sentences and the target answer information.

[0117] Further, in one embodiment, when the current session data corresponds to the third scenario, step M4 includes the following steps M4.1 and M4.2.

[0118] Step M4.1: Determine a preset prompt sentence corresponding to the current session data, and send the preset prompt sentence to the user; Step M4.2: In response to the target information corresponding to the preset prompt statement sent by the user, executing the after-sales processing program associated with the preset prompt statement.

[0119] Specifically, when the system determines that the current session data corresponds to the third scenario, it further determines that the current session data belongs to sub-scenarios such as "replacement of goods", "refund", "complaint", etc., and sends a preset prompt sentence corresponding to the sub-scenarios to the user. After the user provides feedback corresponding to the preset prompt sentence, the system executes the after-sales processing procedure required by the user.

[0120] For example, when a user says "the goods are damaged and need to be returned", the system sends a preset prompt sentence "please provide an unboxing video", and after receiving the unboxing video provided by the user, initiates the return after-sales process. For another example, when a user makes a complaint, the system sends a preset prompt sentence "specify the specific complaint type and provide complaint evidence", and after receiving the complaint information and complaint evidence provided by the user, initiates the complaint handling process.

[0121] Furthermore, based on the above embodiment, after step M4.4, the method further includes the following step M4.3.

[0122] Step M4.3: If the target information corresponding to the preset prompt sentence is not detected in the preset dialogue round, or if it is detected that the current conversation indicates that the user has negative emotions, the call is transferred to manual customer service.

[0123] Specifically, after the system sends a preset prompt statement to the user, if the user fails to provide the target information corresponding to the preset prompt statement in multiple rounds of conversations, the system may have misidentified the user's intention, or the system may be unable to recognize the information currently provided by the user, and manual customer service intervention is required to determine whether to initiate the corresponding after-sales processing procedure.

[0124] For example, the system determines the current conversation as a return scenario and sends a preset prompt sentence to the user: "Please provide an unboxing video." However, the user does not provide an unboxing video in three rounds of conversations. This may be because the system has determined the scenario incorrectly (i.e., the user does not want to return the goods, so no unboxing video will be provided), or the user has provided other evidence to prove that the goods are damaged, but the system cannot recognize it. In this case, manual customer service intervention is required.

[0125] In addition, when the system recognizes that the user has negative emotions, such as the user expressing anger, dissatisfaction or disappointment in the conversation, the system automatically transfers the user to manual customer service.

[0126] Further, in one embodiment, when the current session data corresponds to the second scenario, step M4 includes the following steps M4.4 to M4.7.

[0127] Step M4.4: Obtain basic information, historical purchase behavior data, and browsing behavior data of the user account as basic data, and generate a user profile corresponding to the user account based on the basic data; Step M4.5: adjusting the weights of the knowledge items included in the initial search strategy based on the user portrait to obtain a target search strategy corresponding to the user account; Step M4.6: based on the target detection strategy, retrieving target reply information corresponding to the current session data in the preset knowledge base, and generating customer service information based on the target reply information; Step M4.7: Analyze the feedback behavior of the user account regarding the customer service information, and adjust the target search strategy according to the analysis results. The feedback behavior includes questioning behavior, purchasing behavior, and customer service rating behavior regarding the product information included in the customer service information.

[0128] The applicant found that even if the customer service module of the system can accurately identify the user information after receiving the information sent by the buyer user and generate corresponding customer service information based on the preset knowledge base, the customer service information cannot fully meet the user's information needs. The reason is that the preset knowledge base stores highly standardized target question sentences and target answer information. Therefore, the customer service information generated by the system based on the preset knowledge base also tends to be standardized, and fails to provide personalized customer service information based on the user's actual situation, so that the user needs to make multiple follow-up questions before the needs are met, which seriously affects the efficiency of the customer service process.

[0129] The applicant proposed that for specific users (such as VIP users) or user questions involving complex information processing (such as asking for product recommendations), the system collects user data and builds user portraits, and then adjusts the search strategy of the corresponding preset knowledge base according to the user portrait, so that the final generated customer service information is more in line with the user's actual situation, thereby being able to meet user needs to a greater extent. The essential reason is that the customer service information originally generated is a collection of user questions and preset knowledge bases, but user questions in most cases fail to fully reflect the user's actual situation or user needs, so the system cannot provide personalized customer service information; when the system builds a user portrait, the system can combine the user needs / user characteristics contained in the user portrait that are not reflected in the user's questions into the search strategy (that is, the system takes into account the user's "unspoken needs"), so that the final generated customer service information has personalized characteristics.

[0130] The user data collection process includes: Collect basic information about users, such as age, gender, and region. Basic information can help the system gain a preliminary understanding of the user's background and possible preferences. For example, users of different ages may have different needs and concerns about products. Young users may pay more attention to the fashion of products, while older users may pay more attention to the practicality of products.

[0131] Integrate the user's historical purchase behavior data, including the types of goods purchased, brands, price ranges, purchase frequency, etc. For example, a user who frequently purchases high-end electronic products will be marked in his profile as having high requirements for the quality and performance of electronic products and strong purchasing power.

[0132] Collect users' browsing behavior data, such as the product pages they browse, the duration of stay, search keywords, etc. If users frequently browse products in a certain category, it means that they may have a high interest in that category. For example, if users frequently browse sports equipment pages, the system can infer that they are interested in sports-related products.

[0133] The feature extraction and user profile generation process includes: The system extracts features from the collected user data. For example, it classifies and codes the categories of goods purchased by users, and converts regional information into feature vectors related to culture and consumption habits; and then integrates these feature vectors to generate user portraits through machine learning algorithms, such as clustering algorithms and principal component analysis.

[0134] Generate target search strategies based on user portraits, including: According to the key features in the user portrait, the weights of the knowledge items included in the initial detection strategy are adjusted to generate a target retrieval strategy. For example, if the user portrait shows that the user is an entry-level photography enthusiast, the target retrieval strategy focuses on primary photography equipment, photography effect descriptions, and photography tutorials; if the user portrait shows that the user is an experienced photographer, the target retrieval strategy focuses on advanced photography equipment and the specific parameters of advanced photography equipment. For another example, the target retrieval strategy has a higher allocation weight for knowledge related to brands or product types that users frequently purchase. When users ask questions about skin care, the system prioritizes retrieving knowledge related to the ingredients, usage methods, and applicable skin types of skin care products of brands that users frequently purchase or related brands.

[0135] In addition, since the system's AI module has the possibility of recognizing the probability of user intention, especially user questions involving complex information, the user further proposed to adjust the target retrieval strategy based on the user account's feedback behavior on the customer service information, so that in the event of deviations in previous customer service information, the subsequently generated customer service information can be adjusted in time to meet user needs.

[0136] Specifically, user feedback behavior includes follow-up questions, purchases, and customer service ratings. When a user asks similar questions about the same product included in the customer service information multiple times, or a user does not make a purchase after multiple rounds of dialogue and consultation, or the user's satisfaction rating for the customer service information is low, it means that the customer service information generated by the current target search strategy fails to meet user needs well, and it is necessary to further adjust the target search strategy or transfer to manual customer service.

[0137] In this way, this embodiment ensures the accuracy of customer service information through the preset knowledge base on the basis of setting up a preset knowledge base, and greatly reduces the probability of hallucinations in the AI ​​model; and, the system further adjusts the retrieval strategy corresponding to the preset knowledge base in combination with the user portrait, so that the retrieval results obtained based on the target retrieval strategy can be more in line with the user's actual situation and user needs, thereby achieving the purpose of personalizing customer service information; finally, the system also sets the trigger condition for continuing to optimize the target retrieval strategy, that is, the user's feedback behavior, thereby further improving the communication efficiency between the user and the customer service system.

[0138] Further, based on the above embodiment, after step M4.4, the method further includes the following step M4.8.

[0139] Step M4.8: Based on the user portrait and the target corpus, determine the language style corresponding to the customer service information.

[0140] Specifically, the system analyzes the user's personality characteristics in real time based on the user portrait and target corpus, and then adjusts the interaction style according to the user's personality characteristics to achieve more empathetic and effective communication, thereby meeting the user's personalized interaction needs and providing emotional value to the user.

[0141] For example, when users ask about product specifications and use technical language, the system provides detailed technical instructions, including data tables and data comparison analysis; when users communicate in a spoken style, the system provides simple and direct explanations, sends customer service information in a more life-like style, and even uses local slang.

[0142] The above description is only a preferred embodiment of the present application, and does not limit the patent scope of the present application. All equivalent transformations made by using the contents of the present application specification and drawings under the inventive concept of the present application, or direct / indirect application in the relevant technical field are included in the patent protection scope of the present application.

Claims

1. A customer service information generation method based on classification scenarios, characterized in that: The AI ​​module applied to the system, the method comprises: Step M1: obtaining current session data based on a preset memory window; Step M2: performing a vectorization operation on the current session data to obtain a target corpus; Step M3: calling the AI ​​model to analyze the target corpus, and classifying the current session data into preset scenarios according to the analysis results, wherein the preset scenarios include a first scenario, a second scenario, and a third scenario, wherein the first scenario generates customer service information based on preset configuration information, the second scenario generates customer service information based on a preset knowledge base, and the third scenario generates customer service information based on a preset information collection program; Step M4: Based on the preset scenario, generate customer service information corresponding to the current session data.

2. The method according to claim 1, characterized in that The first scenario includes chat interaction, activity information, process consultation, and logistics inquiry.

3. The method according to claim 1, characterized in that The second scenario includes product consultation.

4. The method according to claim 1, characterized in that The third scenario includes after-sales processing.

5. The method according to claim 1, characterized in that The method further comprises: Step M3.1: Cluster the question sentences in the historical conversation data to obtain corpus clusters; Step M3.2: Determine the corpus cluster whose cluster size reaches a preset size threshold as the target cluster, and determine the target question sentence corresponding to the target cluster; Step M3.3: Determine the target answer information corresponding to the target question sentence in the historical conversation data; Step M3.4: Generate a preset knowledge base based on the target question sentence and the target answer information.

6. The method according to claim 1, characterized in that When the current session data corresponds to the third scenario, step M4 includes: Step M4.1: Determine a preset prompt sentence corresponding to the current session data, and send the preset prompt sentence to the user; Step M4.2: In response to the target information corresponding to the preset prompt statement sent by the user, executing the after-sales processing program associated with the preset prompt statement.

7. The method according to claim 6, characterized in that After step M4.4, the method further comprises: Step M4.3: If the target information corresponding to the preset prompt sentence is not detected in the preset dialogue round, or if it is detected that the current conversation indicates that the user has negative emotions, the call is transferred to manual customer service.

8. The method according to claim 1, characterized in that When the current session data corresponds to the second scenario, step M4 includes: Step M4.4: Obtain basic information, historical purchase behavior data, and browsing behavior data of the user account as basic data, and generate a user profile corresponding to the user account based on the basic data; Step M4.5: adjusting the weights of the knowledge items included in the initial search strategy based on the user portrait to obtain a target search strategy corresponding to the user account; Step M4.6: based on the target detection strategy, retrieving target reply information corresponding to the current session data in the preset knowledge base, and generating customer service information based on the target reply information; Step M4.7: Analyze the feedback behavior of the user account regarding the customer service information, and adjust the target search strategy according to the analysis results. The feedback behavior includes questioning behavior, purchasing behavior, and customer service rating behavior regarding the product information included in the customer service information.

9. The method according to claim 8, characterized in that After step M4.4, the method further comprises: Step M4.8: Based on the user portrait and the target corpus, determine the language style corresponding to the customer service information.

10. A customer service information generation system based on classification scenarios, characterized in that: The system includes an AI module, and the AI ​​module is used to execute the operation instructions included in the customer service information generation method based on classification scenarios as described in any one of claims 1-9.

Images