Dialogue Interaction Methods and Systems Based on Long-History Dialogue Semantic Understanding

Abstract

This invention provides a dialogue interaction method and system based on long-history dialogue semantic understanding, belonging to the field of natural language processing technology. First, it acquires a set of historical dialogue information of target users on an e-commerce platform, including multiple rounds of user inquiries and customer service responses. Then, it performs deep semantic mining on the historical dialogue information set to obtain dialogue intent features and dialogue topic association features. Based on these features, it generates a historical dialogue semantic understanding result that includes a summary of core user needs and a knowledge path for the development of dialogue topics. Based on this historical dialogue semantic understanding result, it generates an appropriate dialogue interaction strategy, including response focus, guidance direction, and rhythm control information. Finally, it executes the strategy to achieve dialogue interaction, outputting response statements and receiving new user input, thereby improving the customer service quality and user experience of the e-commerce platform.

Description

Technical Field

[0001] This invention relates to the field of natural language processing technology, and more specifically, to a dialogue interaction method and system based on long-history dialogue semantic understanding. Background Technology

[0002] In customer service scenarios on e-commerce platforms, conversational interaction serves as a crucial bridge connecting users and the platform. As e-commerce businesses continue to expand and user needs become increasingly diverse, conversations with customer service often involve multiple rounds of complex exchanges, covering aspects such as inquiries about product information, order processing requests, and feedback on after-sales issues.

[0003] However, existing dialogue interaction methods have significant shortcomings in handling users' historical conversations. On the one hand, most methods only grasp the surface information of the conversation, failing to delve into the user's underlying needs and accurately grasp the user's true intentions in different rounds of dialogue. On the other hand, they lack effective analysis and processing of the continuity and transition of topics between different rounds of dialogue, and cannot clearly trace the development of the conversation's themes. This leads customer service representatives to often only provide simple responses to the current problem, failing to offer comprehensive, coherent services that meet the user's underlying needs, resulting in a poor user experience and impacting the service quality and operational efficiency of e-commerce platforms. Summary of the Invention

[0004] In view of the aforementioned problems, and in conjunction with the first aspect of the present invention, a dialogue interaction method based on long-history dialogue semantic understanding is provided, the method comprising:

[0005] Obtain a set of historical dialogue information of target users in an e-commerce platform. The set of historical dialogue information includes multiple rounds of user inquiry statements and corresponding customer service response statements. The user inquiry statements involve product information inquiries, order processing requests, and after-sales problem feedback. The customer service response statements include answers and guidance information for the user inquiry statements.

[0006] Deep semantic mining is performed on the historical dialogue information set to obtain the dialogue intent features and dialogue topic association features of the target user. The dialogue intent features reflect the potential needs of the target user in each round of dialogue, and the dialogue topic association features reflect the continuity and transformation relationship of the topic between different rounds of dialogue.

[0007] Based on the dialogue intent features and the dialogue topic association features, a historical dialogue semantic understanding result is generated, which includes information summarizing the user's core needs and the knowledge path of the dialogue topic development.

[0008] Based on the semantic understanding results of the historical dialogue, an appropriate dialogue interaction strategy is generated. The dialogue interaction strategy includes information on the focus of the response content, guidance direction, and rhythm control method.

[0009] The dialogue interaction strategy is executed to realize dialogue interaction with the target user, output response statements that conform to the dialogue interaction strategy, and receive new input statements from the target user.

[0010] In another aspect, the present invention also provides a dialogue interaction system based on long-history dialogue semantic understanding, including a processor and a machine-readable storage medium connected to the processor. The machine-readable storage medium is used to store programs, instructions or code, and the processor is used to execute the programs, instructions or code in the machine-readable storage medium to implement the above-mentioned method.

[0011] Based on the above, this invention acquires a set of historical dialogue information from e-commerce platforms, including multiple rounds of user inquiries and customer service responses. It then performs deep semantic mining on this historical dialogue information set to obtain the target user's dialogue intent features and dialogue topic association features. This allows for in-depth analysis of the user's potential needs in each round of dialogue and the continuation and transformation of topics between different rounds. The historical dialogue semantic understanding results generated based on the dialogue intent features and dialogue topic association features include information summarizing the user's core needs and the knowledge path of dialogue topic development. This provides a basis for generating an adaptive dialogue interaction strategy based on the historical dialogue semantic understanding results. It clarifies the focus of response content, guidance direction, and rhythm control methods, enabling customer service to provide more comprehensive, coherent services that meet the user's potential needs. Finally, the dialogue interaction strategy is executed to achieve dialogue interaction with the target user, outputting response statements that conform to the strategy and receiving new input statements from the user, effectively improving the customer service quality and user experience of the e-commerce platform. Attached Figure Description

[0012] Figure 1 This is a schematic diagram of the execution flow of the dialogue interaction method based on long history dialogue semantic understanding provided in the embodiments of the present invention.

[0013] Figure 2 This is a schematic diagram of exemplary hardware and software components of a dialogue interaction system based on long-history dialogue semantic understanding provided in an embodiment of the present invention. Detailed Implementation

[0014] The present invention will now be described in detail with reference to the accompanying drawings. Figure 1 This is a flowchart illustrating a dialogue interaction method based on long-history dialogue semantic understanding, provided in one embodiment of the present invention. The following is a detailed description of this dialogue interaction method based on long-history dialogue semantic understanding.

[0015] Step S110: Obtain a set of historical dialogue information of the target user in the e-commerce platform. The set of historical dialogue information includes multiple rounds of user inquiry statements and corresponding customer service response statements. The user inquiry statements involve product information inquiries, order processing requests, and after-sales problem feedback. The customer service response statements include answers and guidance information for the user inquiry statements.

[0016] In this embodiment, the historical dialogue information set of target user B on a comprehensive e-commerce platform is selected as the processing object. This historical dialogue information set covers 12 rounds of dialogue between user B and the platform's customer service within one month. Each round of dialogue includes the inquiry statement made by user B and the response statement made by the customer service to that statement.

[0017] Among these user inquiries, those concerning product information include, "What are the main differences in core performance between the different versions of the multi-functional product mentioned earlier? Also, are there differences in the applicable scenarios for these versions?"; those concerning order processing include, "My order last week, which included multiple product categories, currently shows that some items have been shipped. What is the stock status of the remaining items? Can you expedite the processing to ensure the entire order is shipped as soon as possible?"; and those concerning after-sales feedback include, "One of the items I received had a damaged accessory, and its operating status deviated from the description. How can I return, exchange, or repair it in this situation?"

[0018] The customer service response to product information inquiries states, "The main differences in core performance between different versions of this product lie in operating speed, battery life, and compatibility. The basic version is suitable for simple daily use, the advanced version is better suited for moderate use, and the professional version can meet high-intensity, complex operation scenarios." It also includes guidance: "If you need more detailed comparison parameters, please tell us your main usage needs, and we will provide you with targeted recommendations." Regarding order processing requests, the response is, "Upon checking, the remaining items in your order are currently awaiting shipment. We have forwarded your request to the warehouse and will prioritize processing. Shipment is expected within the next two days, and we will update the logistics information promptly after shipment." For after-sales feedback, customer service responds, "Please first take photos or videos of damaged parts and abnormal operating conditions of the product and upload them to the platform's after-sales channel. Our after-sales specialist will review them within 24 hours. After approval, we will arrange returns, exchanges, or repair services for you based on the specific circumstances, and the platform will bear the related costs."

[0019] Step S120: Perform deep semantic mining on the historical dialogue information set to obtain the dialogue intent features and dialogue topic association features of the target user. The dialogue intent features reflect the potential needs of the target user in each round of dialogue, and the dialogue topic association features reflect the continuity and transformation relationship of topics between different rounds of dialogue.

[0020] In this embodiment, deep semantic mining is performed on the 12-round historical dialogue information set of user B. First, the user's inquiry statements and customer service response statements in each round of dialogue need to be carefully processed to extract key information, and then the user's dialogue intent characteristics are analyzed. These dialogue intent characteristics will clearly show the user's potential needs in each round of dialogue, such as whether they want to learn more about product features, are eager to know the order progress, or are seeking after-sales solutions. At the same time, the topic associations between different rounds of dialogue are also analyzed to determine whether the topic continues or changes, thereby obtaining the dialogue topic association characteristics.

[0021] Step S121: The user inquiry statements and customer service response statements in the historical dialogue information set are segmented according to the natural pauses and semantic integrity of the statements, and multiple dialogue statement units are divided. Each dialogue statement unit retains the original context position information.

[0022] In this embodiment, taking a certain round of dialogue from user B as an example, the user's inquiry is: "Regarding the product with multiple functions mentioned earlier, what are the main differences in core performance between its different versions? Also, are there differences in the applicable scenarios for these versions?". Segmenting according to natural pauses and semantic completeness, this can be divided into two dialogue statement units: the first unit is "Regarding the product with multiple functions mentioned earlier, what are the main differences in core performance between its different versions?", and the second unit is "Also, are there differences in the applicable scenarios for these versions?".

[0023] The customer service response, "The differences in core performance between different versions of this product are mainly reflected in operating speed, battery life, and compatibility. The basic version is suitable for simple daily use, the advanced version is more suitable for moderate use, and the professional version can meet high-intensity, complex operation scenarios. If you need more detailed comparison parameters, please tell us your main usage needs, and we will provide you with targeted recommendations," is divided into three dialogue units: "The differences in core performance between different versions of this product are mainly reflected in operating speed, battery life, and compatibility.", "The basic version is suitable for simple daily use, the advanced version is more suitable for moderate use, and the professional version can meet high-intensity, complex operation scenarios.", and "If you need more detailed comparison parameters, please tell us your main usage needs, and we will provide you with targeted recommendations."

[0024] Each segmented dialogue statement unit retains its position information in the original dialogue, such as which round of dialogue it belongs to, which unit of the user's inquiry statement in that round of dialogue, or which unit of the customer service response statement, etc., so as to facilitate subsequent analysis of contextual semantic relationships.

[0025] Step S122: Extract feature words related to e-commerce transactions from the multiple dialogue statement units, and determine the core feature word set through word frequency statistics and semantic importance assessment. The feature words include product names, transaction process terms, and problem description words.

[0026] In this embodiment, feature words are extracted from all the segmented dialogue statement units. For statement units involving product information, product-related feature words such as "product," "different versions," "core performance," "running speed," "battery life," "compatibility range," "applicable scenarios," "basic version," "advanced version," and "professional version" are extracted; transaction process terms such as "order," "product," "stock status," "processing progress," "shipping," and "logistics information" are extracted from statement units related to order processing; and problem description words such as "product," "accessory parts," "damage," "operating status," "deviation," "return / exchange," and "repair" are extracted from statement units related to after-sales issues.

[0027] Next, we performed word frequency statistics, counting the number of times each feature word appeared in all dialogue sentence units. For example, the word "product" appeared in multiple rounds of dialogue, with a relatively high frequency; "core performance" mainly appeared in dialogues asking for product information, with a medium frequency.

[0028] Then, a semantic importance assessment is performed to evaluate the importance of each feature word in expressing the semantics of the dialogue and user needs. Words such as "core performance," "returns and exchanges," and "logistics information" are directly related to the user's core needs and are therefore of higher importance; while conjunctions such as "in addition" and "among others" are of lower importance.

[0029] Based on the combined results of word frequency statistics and semantic importance assessment, feature words with high importance and frequency that meet certain requirements are selected to form a core feature word set, such as "product", "different versions", "core performance", "running speed", "battery life", "applicable scenarios", "orders", "logistics information", "damage", "returns and exchanges", etc.

[0030] Step S1221: Perform part-of-speech tagging on each dialogue sentence unit to identify nouns, verbs, and adjectives, and then select nouns and verbs that can be used as feature words.

[0031] In this embodiment, natural language processing tools are used to perform part-of-speech tagging on each dialogue statement unit. Taking the dialogue statement unit "What are the main differences in core performance between the different versions of the product with multiple functions mentioned earlier?" as an example, after tagging, "product", "version", and "core performance" are nouns, "mentioned" and "have" are verbs, and "multiple", "different", and "main" are adjectives.

[0032] Nouns and verbs were selected as potential feature words from the annotation results. The nouns selected from the above sentence unit included "product," "version," and "core performance," while the verbs included "mention" and "possess." Considering the e-commerce transaction scenario, "mention" and "possess" have relatively weaker roles in expressing core semantics, while "product," "version," and "core performance" are closely related to product information. Therefore, these nouns were ultimately determined as the feature words for this sentence unit.

[0033] Step S1222: Compare the selected words with the preset e-commerce domain terminology database to match candidate feature words belonging to product names, transaction process terms, and problem description words.

[0034] In this embodiment, the preset e-commerce domain terminology library contains a large number of words related to e-commerce transactions, and the words are classified. For example, the product name category includes various product-related words, the transaction process terminology category covers words related to orders, payments, logistics, etc., and the problem description category includes words such as damage, malfunction, and return.

[0035] The words selected from the dialogue statement units are compared with the vocabulary database. For example, words such as "product", "version", and "core performance" are successfully matched with product name words in the vocabulary database and become candidate feature words for product name; words such as "order", "logistics information", and "shipment" are matched with transaction process terminology words and become candidate feature words for transaction process terminology; words such as "damage", "return", and "repair" are matched with problem description words and become candidate feature words for problem description.

[0036] Step S1223: Count the total number of times each candidate feature word appears in the historical dialogue information set, calculate the distribution frequency of each candidate feature word in each round of dialogue, and obtain the frequency distribution feature.

[0037] In this embodiment, the entire historical dialogue information set is traversed, and the total number of occurrences of each candidate feature word is counted. For example, "product" appears 15 times, "order" appears 10 times, and "return / exchange" appears 8 times.

[0038] Then, the frequency distribution of each candidate feature word in each round of dialogue is calculated, which is the ratio of the number of times the word appears in each round of dialogue to the total number of words in that round. Taking "order" as an example, it appears twice in the 3rd round of dialogue, and the total number of words in that round is 50, so its frequency distribution in the 3rd round is 2 / 50; it appears three times in the 7th round of dialogue, and the total number of words in that round is 60, so its frequency distribution is 3 / 60, and so on, thus obtaining the frequency distribution feature of "order" in each round of dialogue. Other candidate feature words are calculated in the same way to obtain their respective frequency distribution features.

[0039] Step S1224: Based on the semantic weight of each candidate feature word in the sentence, the degree of relevance to the dialogue topic, and the degree of importance in the e-commerce transaction scenario, the semantic importance of the candidate feature words is evaluated, and each candidate feature word is assigned a corresponding importance score.

[0040] In this embodiment, the semantic importance of candidate feature words is evaluated from three dimensions. Regarding semantic weight, it is determined based on the feature word's position in the sentence (e.g., at the beginning, middle, or end) and whether it is emphasized in the sentence; feature words at the beginning of the sentence and those that are emphasized have higher semantic weight. Regarding the degree of relevance to the dialogue topic, if the feature word directly relates to the current dialogue topic, the degree of relevance is high; for example, in a dialogue discussing order issues, "order" and "logistics" are closely related to the topic. Regarding the criticality in e-commerce transaction scenarios, feature words that directly affect the transaction process or core user needs have high criticality, such as "payment" and "returns / exchanges."

[0041] Based on the evaluation results of these three dimensions, each candidate feature word is assigned an importance score between 0 and 1. For example, in a product information consultation dialogue, "core performance" has a high semantic weight, is closely related to the topic, and is crucial for users to understand the product, so it scores 0.85; "in addition" is a conjunction word with a low semantic weight, is not closely related to the topic, and has a low degree of importance, so it scores 0.1.

[0042] Step S1225: Combine frequency distribution features and importance scores to set a screening threshold, select candidate feature words with importance scores higher than the screening threshold to form a core feature word set, and retain the original position mark of each word in the core feature word set in the dialogue sentence unit.

[0043] In this embodiment, a selection threshold is set by comprehensively considering the frequency distribution characteristics and importance scores of candidate feature words. Assuming that the importance score threshold is set to 0.5 after analysis, all candidate feature words are traversed, and words with importance scores higher than 0.5, such as "product," "different versions," "core performance," "running speed," "battery life," "applicable scenarios," "orders," "logistics information," "damage," and "returns," are selected to form the core feature word set.

[0044] Meanwhile, in the core feature word set, each word retains its position mark in the original dialogue statement unit. For example, the word "order" appears in the first unit of the user inquiry statement in the third round of dialogue, and its position mark is "3-U-1", where "3" represents the third round, "U" represents the user inquiry statement, and "1" represents the first unit.

[0045] Step S1226: Deduplicate the core feature word set, and sort the deduplicated core feature word set according to the semantic correlation between the core feature words to determine the correlation ranking result.

[0046] In this embodiment, the core feature word set may contain duplicate words. For example, the word "product" appears in multiple sentence units. After deduplication, only one word "product" is retained.

[0047] Then, the semantic correlation between core feature words is analyzed, measured by methods such as calculating word vector similarity. For example, "core performance" has a high semantic correlation with "running speed" and "battery life," while "order" is closely related to "logistics information" and "shipping." Based on these correlations, the deduplicated set of core feature words is sorted, with words with high correlation grouped together to form a correlation ranking result, such as "product," "different versions," "core performance," "running speed," "battery life," "applicable scenarios," "order," "shipping," "logistics information," "damage," "return," and "repair."

[0048] Step S1227: Optimize the internal structure of the core feature word set through the correlation ranking result to form the final core feature word set.

[0049] In this embodiment, the internal structure of the core feature word set is optimized based on the correlation ranking results. Feature words with high semantic correlation are grouped together to form multiple feature word groups. For example, "product," "different versions," "core performance," "running speed," "battery life," and "applicable scenarios" are grouped into the product information feature word group; "order," "shipping," and "logistics information" are grouped into the order processing feature word group; and "damage," "return," and "repair" are grouped into the after-sales issue feature word group.

[0050] The above optimizations make the internal structure of the core feature word set clearer, which facilitates subsequent semantic analysis and feature extraction. The final core feature word set is presented in the form of these feature word groups.

[0051] Step S123: Perform semantic analysis on the core feature word set in conjunction with the context to determine the basic intent category corresponding to each user's inquiry statement.

[0052] In this embodiment, for each user inquiry statement, semantic analysis is performed on relevant feature words in the core feature word set, taking into account its context, namely, the customer service response statements before and after the statement and other related user inquiry statements. For example, the user inquiry statement "The order I submitted last week, which includes multiple categories, currently shows that some items have been shipped. What is the stock status of the remaining items? Can you speed up the processing to ensure the overall shipment is completed as soon as possible?" has core feature words including "order," "items," "stock status," "processing progress," and "shipped." Considering the context, the user did not mention order-related issues in previous conversations, and this is the first inquiry. The customer service's subsequent response is also about order processing. Therefore, it can be analyzed that the user's intention in this inquiry statement is related to order processing.

[0053] By analyzing each user's inquiry statement as described above, the corresponding basic intent category can be determined, such as product information inquiry, order processing request, after-sales problem feedback, etc.

[0054] Step S1231: Construct an association mapping table between basic intent categories and core feature words. The association mapping table records the mapping relationships between product name feature words corresponding to inquire about product information, transaction process terminology feature words corresponding to inquire about transaction status, and problem description vocabulary feature words corresponding to raise after-sales questions.

[0055] In this embodiment, the constructed association mapping table clarifies the correspondence between different types of core feature words and basic intent categories. Among them, product name feature words such as "product," "different versions," "core performance," and "applicable scenarios" correspond to "inquire about product information" in the basic intent category; transaction process terminology feature words such as "order," "shipment," "logistics information," and "stock status" correspond to "inquire about transaction status"; and problem description terminology feature words such as "damage," "operational status deviation," "return / exchange," and "repair" correspond to "submit after-sales issues."

[0056] This association mapping table is stored in tabular form, which facilitates quick lookup of the basic intent category corresponding to the feature words when analyzing user inquiry statements.

[0057] Step S1232: For each user inquiry statement, extract the core feature words contained in the user inquiry statement from the core feature word set, and preliminarily determine the candidate basic intent category that the user inquiry statement may correspond to based on the mapping relationship recorded in the association mapping table.

[0058] In this embodiment, taking the user inquiry statement "One of the products I received has a damaged accessory, and its operating status is different from the description when used. How should I return or repair it?" as an example, the core feature words extracted from the core feature word set are "product", "accessory", "damaged", "operating status", "deviation", "return" and "repair".

[0059] Based on the association mapping table, these feature words belong to the question description vocabulary category. Therefore, the candidate basic intent category corresponding to the user's inquiry statement is initially determined to be "submitting after-sales questions".

[0060] For user inquiry statements containing multiple types of core feature words, such as product name type and transaction process term type feature words, multiple candidate basic intent categories will be initially identified.

[0061] Step S1233: Analyze the context of the user's inquiry statement based on the mapping logic of the association mapping table, examine the core feature words and semantic tendencies involved in the adjacent customer service response statements and user inquiry statements, and verify whether the initially determined candidate basic intent categories conform to the mapping relationship.

[0062] In this embodiment, taking a user's inquiry statement as an example, the user's inquiry statement contains core feature words such as "product", "core performance" and "order", and the initially determined candidate basic intent categories are "inquiring about product information" and "inquiring about transaction status".

[0063] Analyzing the context, the preceding user inquiries primarily focused on product functionality, and the customer service response also addressed these functionalities. The subsequent response mentioned order processing status. By examining key feature words in the preceding and following statements, such as "functionality" and "parameters" before, and "order number" and "shipping time" after, along with semantic bias, it was found that the statement emphasized further inquiry into product information. Therefore, "inquiring about product information" is the more suitable candidate basic intent category for the mapping relationship.

[0064] Step S1234: If there are multiple candidate basic intent categories, calculate the matching degree between each candidate and the context based on the association strength between the feature words and intent categories in the association mapping table, and select the candidate basic intent category with the highest matching degree as the basic intent category corresponding to the user's inquiry statement.

[0065] In this embodiment, the association mapping table also includes information on the association strength between feature words and intent categories. For example, the association strength between "core performance" and "consult product information" is 0.9, and the association strength between "order" and "inquire about transaction status" is 0.85. For user inquiry statements with multiple candidate basic intent categories, the matching degree between each candidate and the context is calculated.

[0066] Taking user inquiry statements containing core feature words such as "product," "core performance," and "order" as an example, "inquire about product information" and "inquire about transaction status" are candidate basic intent categories. When calculating the matching degree, the association strength between each feature word in the statement and its corresponding intent category is multiplied by the semantic weight of that feature word in the context (the semantic weight is determined based on the frequency and importance of the feature word in the context), and then these products are summed to obtain the matching degree of the candidate basic intent category.

[0067] Assuming the semantic weights of "product" and "core performance" in the context are 0.7 and 0.8 respectively, and the semantic weight of "order" is 0.5, then the matching degree of "inquire about product information" is ("product" association strength × 0.7) + ("core performance" association strength × 0.8), and the matching degree of "inquire about transaction status" is ("order" association strength × 0.5). After calculation, the matching degree of "inquire about product information" is higher than that of "inquire about transaction status," therefore, "inquire about product information" is selected as the basic intent category corresponding to this user's inquiry statement.

[0068] Step S1235: Arrange the basic intent categories corresponding to each user's inquiry statement in the order of the dialogue rounds to form a basic intent category sequence.

[0069] In this embodiment, in user B's 12 rounds of dialogue, each round of user inquiry statements has a corresponding basic intent category, which are as follows in round order: inquire about product information, inquire about product information, inquire about transaction status, inquire about product information, raise after-sales questions, inquire about transaction status, inquire about product information, raise after-sales questions, inquire about transaction status, inquire about product information, raise after-sales questions, inquire about transaction status. These basic intent categories are arranged in order to form a basic intent category sequence.

[0070] Step S1236: Calculate the proportion of each basic intent category in the basic intent category sequence, analyze the distribution characteristics of different intent categories, and the distribution characteristics are positively correlated with the preset association strength of the association mapping table.

[0071] In this embodiment, the frequency of each basic intent category in the basic intent category sequence is counted, and then divided by the total number of rounds (12) to obtain the proportion of each basic intent category. For example, "Inquire about product information" appears 5 times, accounting for 5 / 12; "Inquire about transaction status" appears 4 times, accounting for 4 / 12; and "Submit after-sales questions" appears 3 times, accounting for 3 / 12.

[0072] When analyzing the distribution characteristics, it was found that the basic intent categories with a high proportion also had higher preset association strengths in the association mapping table. For example, the feature words "core performance" and "product" corresponding to "consult product information" had higher association strengths, which is consistent with the positive correlation between distribution characteristics and preset association strengths.

[0073] Step S1237: Adjust the basic intent category sequence based on the distribution characteristics to form the final basic intent category sequence.

[0074] In this embodiment, based on the distribution characteristics, if the proportion of a certain basic intent category is too low and the semantic weight of its corresponding feature word in the context is also low, it may be due to analysis error, and the basic intent category sequence needs to be adjusted.

[0075] For example, the user's inquiry statement in a certain round of dialogue was initially determined to be "raising after-sales issues". However, this category accounts for a very low proportion in the sequence, and the feature word "damaged" in the statement has a low semantic weight in the context. Combining the themes of the preceding and following dialogues, it was found that the statement is more inclined to "inquire about product information". Therefore, the basic intent category of this round was adjusted to "inquire about product information", forming the final basic intent category sequence.

[0076] Step S124: Analyze the topic continuity between different rounds of dialogue statement units based on the basic intent category, and generate topic relevance parameters by calculating topic similarity parameters. The topic relevance parameters are used to represent the degree of topic relevance between adjacent rounds of dialogue.

[0077] In this embodiment, the topic continuity between adjacent dialogue statement units is analyzed based on the basic intent category. If the basic intent categories of two adjacent dialogues are the same, it indicates that the topic may have continuity; if they are different, a topic shift may have occurred. The degree of correlation is quantified by calculating a topic similarity parameter, thereby generating a topic correlation parameter.

[0078] Step S1241: Construct a topic feature vector for each basic intent category. The topic feature vector consists of the core feature words corresponding to the basic intent category and their second weights. The second weights are determined based on the importance score of the feature words.

[0079] In this embodiment, for the basic intent category of "consulting on product information", the corresponding core feature words are "product", "different versions", "core performance", "running speed", "battery life" and "applicable scenarios", with importance scores of 0.8, 0.75, 0.85, 0.7, 0.72 and 0.68, respectively.

[0080] These core feature words are arranged in a certain order, and the importance score corresponding to each feature word is used as its weight to construct a topic feature vector. For example, the topic feature vector of "consulting product information" is [("product", 0.8), ("different versions", 0.75), ("core performance", 0.85), ("running speed", 0.7), ("battery life", 0.72), ("applicable scenarios", 0.68)].

[0081] Similarly, corresponding topic feature vectors are constructed for "inquiring about transaction status" and "submitting after-sales issues". The topic feature vector for "inquiring about transaction status" includes core feature words such as "order", "shipment", and "logistics information" and their weights, while the topic feature vector for "submitting after-sales issues" includes core feature words such as "damage", "return", and "repair" and their weights.

[0082] Step S1242: Select dialogue statement units from adjacent rounds, and extract the topic feature vectors of the basic intent categories corresponding to the dialogue statement units from adjacent rounds respectively.

[0083] In this embodiment, the statement units of the second and third rounds of dialogue are selected. The basic intent category corresponding to the user inquiry statement in the second round is "inquiring about product information", and its topic feature vector V2 is extracted. The basic intent category corresponding to the user inquiry statement in the third round is "inquiring about transaction status", and its topic feature vector V3 is extracted.

[0084] For example, if we select the statement units from the 5th and 6th rounds of dialogue, the basic intent category corresponding to the 5th round is "raising post-sale questions", and the topic feature vector is V5; the basic intent category corresponding to the 6th round is "inquiring about transaction status", and the topic feature vector is V6.

[0085] Step S1243: Calculate the cosine similarity between the two topic feature vectors to obtain the topic similarity parameter. The larger the topic similarity parameter, the closer the topic association between the two rounds of dialogue.

[0086] In this embodiment, when calculating the cosine similarity between topic feature vectors, the core feature words in each topic feature vector are first converted into word vectors. The dimension of the word vectors is determined according to a preset word embedding model, such as 100 dimensions. Then, the word vectors of the core feature words in each topic feature vector are weighted and averaged to obtain the comprehensive vector of the topic feature vector.

[0087] For example, to calculate the cosine similarity between V2 and V3, we first obtain the composite vector A of V2 and the composite vector B of V3. The cosine similarity is calculated by dividing the dot product of A and B by the product of the magnitudes of A and B. The result is the topic similarity parameter. If this parameter is 0.3, it indicates that the topic relevance between the second and third rounds of dialogue is moderate; if the cosine similarity of the topic feature vectors between the first and second rounds of dialogue is 0.8, it indicates that the topics of these two rounds of dialogue are closely related.

[0088] Step S1244: Divide the association level according to the magnitude of the topic similarity parameter. Different association levels correspond to different topic association parameter values.

[0089] In this embodiment, the range of topic similarity parameters is divided into multiple intervals, each interval corresponding to an association level and a topic association parameter value. For example, when the topic similarity parameter is between 0.7 and 1.0, the association level is "closely related" and the topic association parameter value is 0.9; when it is between 0.4 and 0.7, the association level is "generally related" and the topic association parameter value is 0.5; when it is between 0 and 0.4, the association level is "weakly related" and the topic association parameter value is 0.2.

[0090] For example, the topic similarity parameter between the first and second rounds of dialogue is 0.8, corresponding to "close association", and the topic association parameter value is 0.9; the topic similarity parameter between the second and third rounds of dialogue is 0.3, corresponding to "weak association", and the topic association parameter value is 0.2.

[0091] Step S1245: Record the topic relevance parameters of adjacent rounds of dialogue in round order to form a topic relevance sequence. At the same time, calculate the total number of topic transitions in the entire historical dialogue information set and obtain the topic transition frequency by combining the total number of dialogue rounds.

[0092] In this embodiment, the topic relevance parameters of adjacent rounds in the 12 rounds of dialogue are recorded in sequence, such as 0.9 for rounds 1-2, 0.2 for rounds 2-3, 0.3 for rounds 3-4, 0.4 for rounds 4-5, 0.2 for rounds 5-6, 0.5 for rounds 6-7, 0.3 for rounds 7-8, 0.4 for rounds 8-9, 0.2 for rounds 9-10, 0.3 for rounds 10-11, and 0.5 for rounds 11-12, forming a topic relevance sequence [0.9, 0.2, 0.3, 0.4, 0.2, 0.5, 0.3, 0.4, 0.2, 0.3, 0.5].

[0093] The total number of topic transitions was calculated. Each transition was considered a single transition when the underlying intent categories of adjacent rounds differed. The total number of transitions was 7. The total number of rounds in the dialogue was 12, and the topic transition frequency was 7 / 11 (because there are 11 adjacent round pairs in 12 rounds of dialogue).

[0094] Step S1246: Perform sliding window analysis on the topic correlation sequence by setting a fixed-length window, traverse the topic correlation sequence, and calculate the average correlation parameter within the window.

[0095] In this embodiment, the window length is set to 3, meaning each window contains 3 adjacent topic relevance parameters. Starting from the first element of the topic relevance sequence, the window slides. The first window contains [0.9, 0.2, 0.3], and its average relevance parameter is calculated as (0.9 + 0.2 + 0.3) / 3; the second window contains [0.2, 0.3, 0.4], and its average relevance parameter is (0.2 + 0.3 + 0.4) / 3; and so on, until the window slides through the entire sequence, obtaining the average relevance parameter for each window.

[0096] Step S1247: Adjust the topic relevance sequence according to the average relevance parameter, and combine the adjusted topic relevance sequence and topic switching frequency to generate the final topic relevance parameter.

[0097] In this embodiment, if the relevance parameter of a certain topic differs significantly from the average relevance parameter of the window, such as exceeding the ±0.2 range of the average relevance parameter, the parameter is adjusted to bring it closer to the average relevance parameter. For example, if the average relevance parameter of a window is 0.4, and the relevance parameter of a topic within the window is 0.1, the difference is significant, and the parameter is adjusted to 0.3.

[0098] The adjusted topic relevance sequence is smoother and can more accurately reflect the overall trend of topic relevance. Combining the adjusted topic relevance sequence and topic transition frequency, the final topic relevance parameters for each adjacent round of dialogue are determined. For example, for the adjusted parameters, fine-tuning is made according to the frequency of topic transitions. If the frequency of topic transitions is high, some relevance parameters are appropriately reduced to reflect the frequency of topic transitions.

[0099] Step S125: Construct the dialogue intent features and dialogue topic association features of the target user based on the basic intent category and the topic association parameters. The dialogue intent features include an intent category sequence and an intent intensity descriptor. The dialogue topic association features include topic switching frequency and topic duration parameters.

[0100] In this embodiment, the final sequence of basic intent categories is used as the intent category sequence in the dialogue intent features. Simultaneously, an intent intensity descriptor is added to each basic intent category. The intent intensity descriptor is determined based on the semantic weight and frequency of occurrence of that intent category in the dialogue. For example, "inquire about product information" occupies an important position in multiple dialogues, so its intent intensity descriptor is "high intensity"; "raise after-sales questions" appears less frequently, so its intent intensity descriptor is "medium intensity".

[0101] The topic transition frequency in the dialogue topic association feature is 7 / 11 as previously calculated. The topic duration parameter is calculated by counting the number of consecutive rounds of each topic (i.e., basic intent category). For example, if "inquire about product information" appears for 2 consecutive rounds, its duration parameter is 2; if "inquire about transaction status" appears for 1 consecutive round, its duration parameter is 1, etc.

[0102] Step S126: The dialogue intent features are weighted and integrated according to the importance of each round of dialogue in the overall historical dialogue, and a first weight is assigned to each round of dialogue. The round of dialogue with the higher the first weight has a greater impact on the final dialogue intent features.

[0103] In this embodiment, when assessing the importance of each round of dialogue, factors such as the round's position in the dialogue (e.g., newer rounds may be more important), the level of detail in the dialogue content, and whether it involves the user's core needs are considered. Rounds with higher importance are assigned a higher first weight; for example, the 10th round of dialogue involves the user's final confirmation of the product, so the first weight is 0.9. Earlier rounds have a lower first weight; for example, the first weight of the 1st round is 0.5.

[0104] Multiply the intent category sequence elements of each round of dialogue by their corresponding first weight to obtain a weighted intent category sequence, thereby achieving weighted integration of dialogue intent features and making the final dialogue intent features more reflective of the intent of important rounds.

[0105] Step S127: Perform temporal calibration on the dialogue topic association features. Cross-validate the weighted and integrated dialogue intent features and the temporally calibrated dialogue topic association features. If the target user dialogue tendencies reflected by the two are consistent, then the dialogue intent features and the dialogue topic association features are confirmed to be obtained. If the target user dialogue tendencies reflected by the two are different, then readjust the feature extraction parameters and perform deep semantic mining again until the dialogue intent features and the dialogue topic association features are obtained.

[0106] In this embodiment, the topic association features of the dialogue are calibrated in a temporal manner. That is, the time stamps of the topic duration parameter and the topic association parameter are adjusted according to the time sequence of the dialogue rounds to ensure that they are consistent with the timeline of the actual dialogue.

[0107] During cross-validation, examine whether the frequently occurring intent categories in the weighted and integrated dialogue intent features match the topics with longer durations and higher relevance in the dialogue topic association features. For example, if "inquiring about product information" is a high-frequency, high-intensity intent in the dialogue intent features, and the corresponding topic in the dialogue topic association features has a longer duration and higher relevance, then the two reflect consistent dialogue tendencies, confirming the final dialogue intent features and dialogue topic association features.

[0108] If there is a discrepancy, such as the dialogue intent feature showing "inquiring about the transaction status" as the main intent, but the duration of the ongoing topic in the dialogue topic association feature is short and the association is low, then it is necessary to readjust the importance scoring parameters and the weights of the topic feature vectors during feature extraction, and perform deep semantic mining again until the two tend to be consistent.

[0109] Step S130: Generate historical dialogue semantic understanding results based on the dialogue intent features and the dialogue topic association features. The historical dialogue semantic understanding results include user core needs summary information and dialogue topic development knowledge path.

[0110] In this embodiment, based on the intent category sequence and intent intensity descriptor in the dialogue intent features, the core needs of the user in the entire historical dialogue are summarized; combined with the topic switching frequency, topic duration parameter and topic relevance parameter in the dialogue topic association features, a knowledge path for the development of dialogue topics is constructed, and these two parts are integrated to form the semantic understanding result of historical dialogue.

[0111] Step S131: Perform statistical analysis on the intent category sequence in the dialogue intent features, identify the intent category with the highest frequency, and determine the main intent direction of the target user by combining the intent intensity descriptor.

[0112] In this embodiment, the intent category sequence was statistically analyzed. "Inquire about product information" appeared 5 times, "Inquire about transaction status" appeared 4 times, and "Submit after-sales questions" appeared 3 times. Therefore, "Inquire about product information" is the intent category with the highest frequency. Its intent strength descriptor is "high intensity". Based on the analysis, it was determined that the target user's main intent direction is to gain a deeper understanding of product-related information in order to make a purchase decision.

[0113] Step S132: Based on the main intent direction, integrate the relevant user inquiry statements and customer service response statements, extract the key information, and summarize the core needs of the target user in the entire historical dialogue.

[0114] In this embodiment, all relevant user inquiry statements and customer service response statements are integrated around the main intent of "inquiring about product information". The user inquiry statements involve information such as the core performance and applicable scenarios of different product versions, while the customer service response statements include the differences between the versions in terms of running speed, battery life, etc.

[0115] Extract key information from these statements, such as the core performance differences between different versions and the division of applicable scenarios, and summarize the core needs of the target users as: to clearly understand the core performance and applicable scenarios of different versions of the product in order to choose the version that best suits their needs.

[0116] Step S133: Analyze the topic association sequence and topic switching frequency in the dialogue topic association features, and analyze the occurrence order, duration and switching nodes of different topics in the dialogue.

[0117] In this embodiment, the order of occurrence of different topics is analyzed by combining parameters such as topic relevance sequence, topic switching frequency, and topic duration: Consulting product information—Consulting product information—Inquiring about transaction status—Consulting product information—Raising after-sales questions—Inquiring about transaction status—Consulting product information—Raising after-sales questions—Inquiring about transaction status—Consulting product information—Raising after-sales questions—Inquiring about transaction status.

[0118] The duration of each topic is determined by the number of consecutive rounds it appears. "Inquire about product information" can appear for a maximum of 2 consecutive rounds, with a duration of 2; "Inquire about transaction status" can appear for 1 consecutive round, with a duration of 1; "Submit after-sales questions" can appear for 1 consecutive round, with a duration of 1.

[0119] Theme transition nodes are the locations where the themes of adjacent rounds differ, such as the transition between rounds 2-3, 3-4, and 4-5.

[0120] Step S1331: Traverse the topic relevance sequence and determine consecutive significantly related dialogue segments based on the topic relevance parameter values. Each significantly related dialogue segment corresponds to a continuous topic.

[0121] In this embodiment, a topic relevance parameter value greater than 0.6 is defined as a significant relevance. The topic relevance sequence is traversed to find consecutive segments with parameter values ​​greater than 0.6. For example, the topic relevance parameter values ​​for rounds 1-2 are 0.7 and 0.8, for round 3 it is 0.5, and for rounds 4-6 it is 0.7, 0.65, and 0.72. Therefore, rounds 1-2 and 4-6 can be identified as consecutive significantly related dialogue segments. Specifically, rounds 1-2 revolve around the basic functions of the product, with the corresponding continuous topic being "consultation on basic product functions"; rounds 4-6 focus on the usage methods of the product, with the corresponding continuous topic being "discussion on product usage methods".

[0122] Step S1332: Record the start and end rounds of each significantly related dialogue segment in the historical dialogue, calculate the difference between the end round and the start round, and obtain the duration of the ongoing topic.

[0123] In this embodiment, for the significantly related dialogue segments in rounds 1-2, the starting round is 1 and the ending round is 2. The difference between the ending round and the starting round is calculated as 2-1=1, meaning the duration of the topic "Consultation on Basic Product Functions" is one round interval. For the significantly related dialogue segments in rounds 4-6, the starting round is 4 and the ending round is 6, with a difference of 6-4=2. The duration of the topic "Discussion on Product Usage Methods" is two round intervals. All significantly related dialogue segments are processed in the same way, and their starting and ending rounds are recorded, with the duration of each topic calculated.

[0124] Step S1333: Determine the order in which different topics appear in the historical dialogue based on the starting round order of each significantly related dialogue segment, and arrange the topic names in order.

[0125] In this embodiment, the starting rounds of each significantly related dialogue segment are 1, 4, 8, and 11, respectively. Following the order of the starting rounds from earliest to latest, the corresponding topics are "Inquiry about basic product functions," "Discussion on product usage methods," "Order and logistics tracking," and "After-sales problem resolution." Therefore, the order in which these topics appear in the historical dialogues is determined to be "Inquiry about basic product functions"—"Discussion on product usage methods"—"Order and logistics tracking"—"After-sales problem resolution," and the topic names are arranged in this order.

[0126] Step S1334: Locate the position in the topic relevance sequence where the relevance parameter value suddenly decreases, and use it as a potential conversion node for topic conversion.

[0127] In this embodiment, the topic relevance sequence is [0.7, 0.8, 0.5, 0.7, 0.65, 0.72, 0.4, 0.68, 0.7, 0.55, 0.69, 0.71]. Observation reveals that at round 3, the relevance parameter value suddenly decreases from 0.8 in round 2 to 0.5; at round 7, it suddenly decreases from 0.72 in round 6 to 0.4; and at round 10, it suddenly decreases from 0.7 in round 9 to 0.55. These locations where parameter values ​​suddenly decrease, namely rounds 3, 7, and 10, are marked as potential topic transition nodes.

[0128] Step S1335: Verify the rationality of the potential conversion node by combining the topic conversion frequency. If it is confirmed to be rational, mark it as a formal conversion node and determine the corresponding topics before and after the node.

[0129] In this embodiment, statistics show that the topic transition frequency throughout the entire historical dialogue is 3 times. There are 3 potential transition nodes in rounds 3, 7, and 10, consistent with the topic transition frequency, indicating the rationality of these potential transition nodes. After marking them as official transition nodes, the topic before round 3 is determined to be "Consultation on basic product functions," and the topic after round 7 is "Discussion on product usage methods"; the topic before round 7 is "Discussion on product usage methods," and the topic after round 10 is "Order and logistics tracking"; the topic before round 10 is "Order and logistics tracking," and the topic after round 10 is "After-sales problem resolution."

[0130] Step S1336: Organize the order of appearance, duration and transformation nodes of different topics into structured data.

[0131] In this embodiment, the following are organized: the order of appearance of the topic "Consultation on Basic Product Functions" (1), the duration of one round, and the position before the corresponding conversion node (none, as it is the first topic); the order of appearance of the topic "Discussion on Product Usage Methods" (2), the duration of two rounds, and the conversion node of round 3; the order of appearance of the topic "Order Logistics Inquiry" (3), the duration of two rounds (rounds 8-9), and the conversion node of round 7; and the order of appearance of the topic "After-Sales Problem Solving" (4), the duration of two rounds (rounds 11-12), and the conversion node of round 10. This forms structured data, which is presented in list form. Each element contains the topic name, the order of appearance, the duration, and the conversion node information.

[0132] Step S1337: Normalize the duration of each topic, convert the duration into a proportion of the total dialogue duration, and adjust the duration parameter in the structured data according to the normalization result.

[0133] In this embodiment, the entire dialogue consists of 12 rounds, with a total duration of 11 intervals. The duration of the topic "Consultation on Basic Product Functions" is 1 round interval, which is 1 / 11 after normalization; the duration of "Discussion on Product Usage Methods" is 2 round intervals, which is 2 / 11 after normalization; the duration of "Order Logistics Inquiry" is 2 round intervals, which is 2 / 11 after normalization; and the duration of "After-Sales Problem Solving" is 2 round intervals, which is 2 / 11 after normalization. Based on these normalization results, the duration parameters of each topic in the structured data are adjusted to the corresponding proportional values.

[0134] Step S134: Construct the thematic development knowledge path of the entire historical dialogue based on the order of appearance of the topics, duration of duration, and transition nodes.

[0135] In this embodiment, the order in which the topics appear serves as the timeline, and the duration and transition nodes of each topic are marked in the path. The starting topic is "Inquire about product information," which continues for 2 rounds and then switches to "Inquire about transaction status" at the transition node of round 2-3. After this topic continues for 1 round, it switches back to "Inquire about product information" at the transition node of round 3-4. And so on, recording the transition nodes and duration of each topic in sequence to form a complete topic development knowledge path: Inquire about product information (rounds 1-2) — Inquire about transaction status (round 3) — Inquire about product information (round 4) — Submit after-sales questions (round 5) — Inquire about transaction status (round 6) — Inquire about product information (round 7) — Submit after-sales questions (round 8) — Inquire about transaction status (round 9) — Inquire about product information (round 10) — Submit after-sales questions (round 11) — Inquire about transaction status (round 12).

[0136] Step S135: Integrate the user's core needs summary information and the knowledge path of the dialogue topic development to generate the semantic understanding results of the historical dialogue.

[0137] In this embodiment, the user's core needs summary information—"clarifying the core performance and applicable scenarios of different product versions and selecting the version suitable for their own needs"—is integrated with the topic development knowledge path. The topic stages related to the core needs are marked in the knowledge path; for example, the "consulting about product information" stage corresponds to the main exploration process of the core needs, and the "raising after-sales questions" stage corresponds to the usage guarantee stage extending the needs, thus initially forming the semantic understanding results of historical dialogues.

[0138] Step S136: The user core needs summary information is hierarchically divided, the core needs are decomposed into multi-level needs, the subordinate relationship between each level of needs is determined, and the hierarchical division result is obtained.

[0139] In this embodiment, the core requirement of "clarifying the core performance and applicable scenarios of different product versions and selecting the version suitable for one's own needs" is divided into three levels of requirements: Level 1 requirement is "product selection decision support"; Level 2 requirements are "core performance comparison" and "applicable scenario matching"; Level 3 requirements are "differences in operating speed," "comparison of battery life," "adaptability to daily scenarios," and "adaptability to high-intensity scenarios," etc. The subordinate relationship of each level of requirements is as follows: Level 3 requirements support Level 2 requirements, and Level 2 requirements serve Level 1 requirements, forming the hierarchical division result.

[0140] Step S137: Divide the topic development knowledge path into stages according to the topic transition nodes, divide the entire dialogue process into different stages, and each stage corresponds to one or more closely related topics to obtain the stage division results.

[0141] In this embodiment, the entire dialogue process is divided into four stages, with the topic transition node as the dividing point: Stage 1 (rounds 1-2) corresponds to the topic of "inquiring about product information," which is the initial demand exploration stage; Stage 2 (rounds 3-4) includes the topics of "inquiring about transaction status" and "inquiring about product information," which is the demand and process parallel stage; Stage 3 (rounds 5-8) includes the topics of "raising after-sales issues," "inquiring about transaction status," and "inquiring about product information," which is the demand deepening and issue feedback stage; Stage 4 (rounds 9-12) includes the topics of "inquiring about transaction status," "inquiring about product information," and "raising after-sales issues," which is the decision confirmation and assurance stage, thus obtaining the stage division result.

[0142] Step S138: Optimize the historical dialogue semantic understanding result based on the hierarchical division result and the stage division result to form the final historical dialogue semantic understanding result.

[0143] In this embodiment, based on the hierarchical division results, the specific content and relationships of each level of requirements are supplemented into the historical dialogue semantic understanding results. Combined with the stage division results, the core requirement level corresponding to each stage is marked in the topic development knowledge path. For example, stage 1 corresponds to the preliminary exploration of the second-level requirement "core performance comparison," and stage 3 corresponds to the in-depth confirmation of the third-level requirement "differences in running speed." Through the above optimizations, the historical dialogue semantic understanding results clearly present the hierarchical structure of user requirements and clarify the correspondence between topic development and requirement evolution, forming the final historical dialogue semantic understanding results.

[0144] Step S140: Generate an adapted dialogue interaction strategy based on the semantic understanding results of the historical dialogue. The dialogue interaction strategy includes information on the focus of the response content, guidance direction, and rhythm control method.

[0145] In this embodiment, a dialogue interaction strategy is generated based on the user's core needs summary information and the dialogue topic development knowledge path obtained from the semantic understanding results of historical dialogues. The user's core needs summary information shows that the user is mainly concerned with product functions, order logistics, and after-sales issues. The dialogue topic development knowledge path presents the order and transition of topics. Based on this, the response content should focus on detailed explanations of product functions, real-time status of order logistics, and solutions to after-sales issues; the guidance direction should gradually transition from product function introduction to order follow-up, and then to after-sales support; the pacing control method should maintain a moderate response speed and sentence length based on the user's previous dialogue habits.

[0146] Step S141: Analyze the user's core needs summary information in the semantic understanding results of the historical dialogue, determine the information domain covered by the response content, and determine the focus information of the response content.

[0147] In this embodiment, the core user needs are analyzed and summarized, revealing that these needs include understanding the various functional features of the product, tracking the order's logistics progress, and resolving problems encountered during product use. Therefore, the response content needs to cover the product functionality, order logistics, and after-sales issues. Among these areas, users inquire most frequently and in detail about product functions and have the most urgent need for after-sales solutions. Therefore, product functionality descriptions and after-sales solutions are prioritized in the response content, while order logistics information is a secondary focus.

[0148] Step S142: Analyze the knowledge path of the dialogue topic, determine the current stage of development and reference extension direction of the dialogue topic, and determine the direction of dialogue guidance in combination with the user's core needs.

[0149] In this embodiment, the knowledge path of the dialogue topic shows that the current dialogue topic is in the "discussion on how to use the product" stage, having previously gone through the "consultation on basic product functions" stage, and may extend to the "order and logistics inquiry" stage. Considering the user's core needs, after understanding how to use the product, the user is likely to be concerned about when the order will be delivered. Therefore, the direction of the dialogue is determined to be a gradual shift from a detailed explanation of how to use the product to informing the user of order and logistics information, while also addressing any after-sales questions the user may raise during the guidance process.

[0150] Step S143: Referencing the target user's inquiry frequency, inquiry sentence length, and response speed in historical dialogues, assess the target user's dialogue rhythm habits and determine the rhythm control method information to be adopted in new dialogue interactions. The rhythm control method information includes control information for response speed and response sentence length.

[0151] In this embodiment, statistical analysis of historical dialogues revealed that the target user sends an inquiry message on average every 20 minutes, indicating a moderate inquiry frequency; the average length of each inquiry message is 25 Chinese characters, a moderate length; and the average time from receiving a customer service response to sending the next inquiry is 15 minutes, indicating a moderate response speed. Based on these data, the target user's dialogue rhythm is assessed as moderate. Therefore, in new dialogue interactions, the response speed should be controlled within 5-10 minutes, and the response message length should be controlled between 30-50 Chinese characters, forming the corresponding rhythm control information.

[0152] For example, in step S1431, the number of user inquiry statements sent by the target user within a unit of time in the historical dialogue is counted to obtain the inquiry frequency of the target user.

[0153] In this embodiment, historical conversations from the past 2 hours are selected as the statistical time period. During this time period, the target user sent a total of 6 inquiry messages. The inquiry frequency of the target user is calculated as 6 messages / 2 hours = 3 messages / hour, that is, the target user sends an average of 3 inquiry messages per hour, which is used as the inquiry frequency data of the target user.

[0154] Step S1432: Measure the number of words in each user inquiry statement from the target user in the historical dialogue and calculate the average statement length.

[0155] In this embodiment, the number of characters in each inquiry from the target user was measured, which were 22, 28, 24, 30, 20, and 26 characters respectively. These character counts were added together, totaling 22 + 28 + 24 + 30 + 20 + 26 = 150 characters. Dividing this by the number of inquiry statements (6) yielded an average statement length of 150 / 6 = 25 characters.

[0156] Step S1433: Calculate the time interval between the target user receiving the customer service response statement and sending the next user inquiry statement in the historical dialogue to obtain the target user's response speed.

[0157] In this embodiment, the time interval between each time the target user receives a customer service response and sends the next inquiry is recorded as 12 minutes, 18 minutes, 14 minutes, 16 minutes, 13 minutes, and 17 minutes. These time intervals are added together, totaling 12 + 18 + 14 + 16 + 13 + 17 = 90 minutes. Dividing this by the number of intervals (6), the average time interval is 90 / 6 = 15 minutes, meaning the target user's response time is an average of once every 15 minutes.

[0158] Step S1434: Construct a user dialogue rhythm evaluation model by combining the consultation frequency, the average sentence length, and the response speed, and classify the dialogue rhythm habits of target users into different types.

[0159] In this embodiment, the constructed user dialogue rhythm evaluation model includes three input dimensions: consultation frequency, average sentence length, and response speed. Consultation frequency is set as high (>5 messages / hour), medium (3-5 messages / hour), and low (<3 messages / hour); average sentence length is long (>40 characters), medium (20-40 characters), and short (<20 characters); response speed is fast (<10 minutes), medium (10-20 minutes), and slow (>20 minutes). The target user's consultation frequency of 3 messages / hour (medium), average sentence length of 25 characters (medium), and response speed of 15 minutes (medium) are input into the user dialogue rhythm evaluation model. The model outputs the target user's dialogue rhythm habit type as "medium rhythm type."

[0160] Step S1435: Preset corresponding rhythm control parameters for different dialogue rhythm types. The rhythm control parameters include the fast and slow range of response speed and the appropriate range of response sentence length.

[0161] In this embodiment, for "fast-paced" users, the preset response speed range is 1-5 minutes, and the appropriate response sentence length is 10-20 words; for "medium-paced" users, the response speed range is 5-10 minutes, and the appropriate response sentence length is 30-50 words; for "slow-paced" users, the response speed range is 10-15 minutes, and the appropriate response sentence length is 50-80 words. These preset pace control parameters are stored in the system's parameter library for easy retrieval based on the user's pace type.

[0162] Step S1436: Select the corresponding rhythm control parameters according to the dialogue rhythm type of the target user, and determine the rhythm control method information to be used in the new dialogue interaction.

[0163] In this embodiment, the target user is classified as "medium-paced." Corresponding pacing control parameters are selected from the parameter library, namely a response speed range of 5-10 minutes and a suitable response sentence length of 30-50 words. These parameters are determined as the pacing control method to be used in new dialogue interactions, ensuring that the response matches the user's pacing habits.

[0164] Step S1437: Compare and analyze the dialogue rhythm habits of target users at different topic stages to identify stage differences in rhythm habits.

[0165] In this embodiment, the frequency, average sentence length, and response speed of user inquiries were statistically analyzed across four thematic stages: "Basic Product Function Consultation," "Discussion on Product Usage Methods," "Order and Logistics Inquiry," and "After-Sales Problem Solving." It was found that in the "Order and Logistics Inquiry" stage, the user inquiry frequency was 4 inquiries per hour, the average sentence length was 20 characters, and the response time was 10 minutes, indicating a relatively fast pace. In the "Discussion on Product Usage Methods" stage, the inquiry frequency was 2 inquiries per hour, the average sentence length was 30 characters, and the response time was 20 minutes, indicating a relatively slow pace. Comparative analysis revealed these differences in pace and habits across the stages.

[0166] Step S1438: Adjust the rhythm control method information according to the stage difference information to form the final rhythm control method information.

[0167] In this embodiment, based on the stage difference information, in the "Order Logistics Inquiry" stage, the rhythm control method information is adjusted to a response speed range of 3-7 minutes and a suitable response sentence length range of 20-40 words; in the "Discussion on Product Usage Methods" stage, it is adjusted to a response speed range of 7-12 minutes and a suitable response sentence length range of 40-60 words; other stages maintain the original "medium rhythm" parameters. After the above adjustments, the final rhythm control method information is formed.

[0168] Step S144: Construct a corresponding dialogue interaction strategy based on the response content focus information, the guidance direction, and the rhythm control method information; set priorities for the response content focus information according to different user core need levels; adjust the intensity of the guidance direction according to different stages of topic development; and optimize the dialogue interaction strategy based on the set priority and intensity adjustment results to form the final dialogue interaction strategy.

[0169] In this embodiment, a preliminary dialogue interaction strategy is first constructed based on the focus of the response content (primarily product feature introduction and after-sales problem solutions, with order logistics information as secondary), the direction of guidance (from product usage methods to order logistics, while also considering after-sales service), and the pacing control method. Then, priorities are assigned to the user's core needs: "Understanding Product Features" and "Solving After-Sales Problems" have high priority, while "Order Logistics Inquiry" has medium priority. Depending on the stage of the topic development, the intensity of guidance to order logistics is set to medium during the "Discussion on Product Usage Methods" stage, and increases to high after entering the "Order Logistics Inquiry" stage. Based on these priority and intensity adjustments, the preliminary dialogue interaction strategy is optimized, clarifying the response focus, guidance intensity, and pacing control details at different stages, thus forming the final dialogue interaction strategy.

[0170] Step S150: Execute the dialogue interaction strategy to realize dialogue interaction with the target user, output response statements that conform to the dialogue interaction strategy, and receive new input statements from the target user.

[0171] In this embodiment, the operation is executed according to the final dialogue interaction strategy. When a user inquires about a specific detail of product usage, the system explains that detail in detail based on the focus of the response, while responding within 7 minutes according to a pacing control method. The message length is controlled to around 40 characters, and the system appropriately guides the user to pay attention to upcoming logistics information. After outputting the response, the system waits in real time for new input from the user. Once received, it is incorporated into a new dialogue flow, and the interaction continues according to the strategy.

[0172] Step S151: Select a matching response template from the preset response template library according to the response content focus information in the dialogue interaction strategy, and fill in information related to the user's core needs and guiding statements corresponding to the guidance direction to generate an initial response statement.

[0173] In this embodiment, the response focuses on product feature introduction and after-sales problem solutions. From a pre-set response template library, the template related to product feature introduction, "Regarding [product features], its main features include [Feature 1] and [Feature 2], which can play a good role in [usage scenarios]. If you would like to know more about [related features], please let us know at any time.", and the template related to after-sales problem solutions, "Regarding the [after-sales problem] you mentioned, our solution is [solution steps], which can resolve the [problem effect] after execution. We will notify you of any subsequent progress in a timely manner.", are selected.

[0174] When a user inquires about a product's function, fill in relevant product function information, such as, "Regarding the product's energy-saving mode, its main features include low-power operation and automatic performance adjustment, which can play a good role in long-term use scenarios. If you would like to know how to switch to this mode, please let us know at any time." Also, add a guiding statement, "In addition, the logistics information for your purchased product is expected to be updated today, so please keep an eye out for it." to generate an initial response statement.

[0175] Step S152: Adjust the initial response statement according to the rhythm control method information, and output the adjusted response statement.

[0176] In this embodiment, the rhythm control information requires that during the "Discussion on Product Usage Methods" phase, the response statement length be 40-60 characters, and the response speed be within 7-12 minutes. The generated initial response statement length was 70 characters, exceeding the range, and needed to be adjusted. Part of the description was simplified to approximately 50 characters, such as, "Regarding the product's energy-saving mode, its characteristics are low power consumption and automatic adjustment, suitable for long-term use. Please let us know if you would like to know how to switch modes. Logistics information is expected to be updated today, please pay attention." After confirming that the adjusted statement meets the rhythm control requirements, the response statement is output within 10 minutes.

[0177] Step S153: After outputting the adjusted response statement, monitor the target user's input status in real time, wait to receive new input statements from the target user, and record the content and receiving time of the new input statements from the target user.

[0178] In this embodiment, after the adjusted response statement is output, the user's input interface can be monitored in real time and kept in standby mode. When the user enters a new statement "How do I switch to energy-saving mode? Can I check the logistics information now?" after 30 minutes, the statement is immediately received and its content is recorded as "How do I switch to energy-saving mode? Can I check the logistics information now?", with the reception time being 2:15 PM that day.

[0179] Step S154: Add the new input statement of the target user to the historical dialogue information set.

[0180] In this embodiment, the newly entered statement "How do I turn on energy-saving mode?" is added to the historical dialogue information set. During the addition process, the time of receipt of the statement and its position in the dialogue round can be automatically recorded. Since the statement was issued by the user in the 13th round of the dialogue, it will be marked as the 13th round user inquiry statement and associated with the corresponding customer service response statement.

[0181] Meanwhile, to ensure the integrity and consistency of the historical dialogue information set, the format of the added set can be validated to ensure that the storage format of the statements is consistent with the previous dialogue statements, including the format of statement content, location markers, timestamps, and other information. Furthermore, considering that the content of this statement may be related to previous product function inquiries, the association information between this statement and related topics can be updated in the internal index to facilitate rapid location and correlation analysis during subsequent deep semantic mining.

[0182] Step S155: By analyzing the target user's response speed after receiving the adjusted response statement and the content of the target user's new input statement, the effect of the adjusted response statement is evaluated. If the evaluation result shows that the response statement does not meet the target user's expectations, the focus information of the response content or the rhythm control method information in the dialogue interaction strategy is adjusted, and the response statement is regenerated.

[0183] In this embodiment, after receiving the adjusted response, the target user sent a new input statement, "How do I turn on power saving mode?", one minute later. First, the response speed is analyzed by comparing this response time with the user's average response time in historical conversations. If the historical average response time is two minutes, this one-minute response time is faster than the average, indicating that the user responded quickly to the response statement.

[0184] Next, we analyzed the new input statement, which was a question about how to activate the energy-saving mode. The previously adjusted response mainly focused on the product's basic functions and did not address the energy-saving mode. Considering both the response speed and the statement content, we determined that while the response elicited a quick user feedback, it did not fully address the user's underlying needs. Therefore, the effectiveness evaluation result was that it did not meet the target user's expectations.

[0185] Based on this, the focus of the response content in the dialogue interaction strategy can be adjusted to include information related to the energy-saving mode. At the same time, considering the user's relatively quick response, the current pace control method should be maintained. Then, the response statement is regenerated according to the adjusted strategy, such as, "The energy-saving mode of this product is activated by pressing and holding the function key for 3 seconds. The screen will display 'Energy Saving' to indicate successful activation. Please feel free to contact us if you have any other questions."

[0186] Step S156: Output the regenerated response statement to the target user, and receive new input statements from the target user again until the dialogue interaction is completed.

[0187] In this embodiment, the regenerated response statement "The energy-saving mode of this product is activated by pressing and holding the function key for 3 seconds. The screen will display the word 'Energy Saving' to indicate successful activation. If you have any other operational questions, please feel free to contact us." will be output to the target user.

[0188] The system continuously monitors the user's input status and waits to receive new input statements. If the user subsequently inputs "I understand, thank you," it indicates that the user's problem has been resolved and the dialogue interaction is complete. If the user continues to ask other questions, such as "How much will the battery life be extended in energy-saving mode?", the above steps can be repeated to generate, output, and receive response statements again until the user has no further input or explicitly indicates the end of the dialogue.

[0189] Throughout the process, the historical dialogue information set is continuously updated to ensure that every interaction is accurately recorded. Simultaneously, all data related to user dialogues is encrypted and stored, and data anonymization techniques are used to process user identifiers and other private information to prevent privacy leaks and protect user data security.

[0190] Figure 2 The illustration shows exemplary hardware and software components of a dialogue interaction system 100 based on long-history dialogue semantic understanding, which can implement the ideas of this application, according to some embodiments of this application. For example, processor 120 can be used in the dialogue interaction system 100 based on long-history dialogue semantic understanding and to perform the functions in this application.

[0191] For example, a dialogue interaction system 100 based on long-history dialogue semantic understanding may include a network port 110 connected to a network, one or more processors 120 for executing program instructions, a communication bus 130, and various forms of storage media 140, such as a disk, ROM, or RAM, or any combination thereof. Exemplarily, the dialogue interaction system 100 based on long-history dialogue semantic understanding may also include program instructions stored in ROM, RAM, or other types of non-transitory storage media, or any combination thereof. The methods of this application can be implemented according to these program instructions. The dialogue interaction system 100 based on long-history dialogue semantic understanding also includes an I / O interface 150 between the computer and other input / output devices.

[0192] Furthermore, this embodiment of the invention also provides a readable storage medium, wherein computer-executable instructions are preset in the readable storage medium, and when the processor executes the computer-executable instructions, the above-mentioned dialogue interaction method based on long history dialogue semantic understanding is implemented.

[0193] It should be noted that, in order to simplify the description of the present invention and thus help to understand one or more embodiments of the invention, multiple features may sometimes be grouped into one embodiment, drawing or description thereof in the foregoing description of the embodiments of the present invention.

Claims

1. A dialogue interaction method based on long-history dialogue semantic understanding, characterized in that, The method includes: Obtain a set of historical dialogue information of target users in an e-commerce platform. The set of historical dialogue information includes multiple rounds of user inquiry statements and corresponding customer service response statements. The user inquiry statements involve product information inquiries, order processing requests, and after-sales problem feedback. The customer service response statements include answers and guidance information for the user inquiry statements. Deep semantic mining is performed on the historical dialogue information set to obtain the dialogue intent features and dialogue topic association features of the target user. The dialogue intent features reflect the potential needs of the target user in each round of dialogue, and the dialogue topic association features reflect the continuity and transformation relationship of the topic between different rounds of dialogue. Based on the dialogue intent features and the dialogue topic association features, a historical dialogue semantic understanding result is generated, which includes information summarizing the user's core needs and the knowledge path of the dialogue topic development. Based on the semantic understanding results of the historical dialogue, an appropriate dialogue interaction strategy is generated, which includes information on the focus of the response content, the direction of guidance, and the rhythm control method. The dialogue interaction strategy is executed to realize dialogue interaction with the target user, output response statements that conform to the dialogue interaction strategy, and receive new input statements from the target user. The step of generating historical dialogue semantic understanding results based on the dialogue intent features and the dialogue topic association features includes: Statistical analysis is performed on the intent category sequence in the dialogue intent features to identify the intent category with the highest frequency of occurrence, and the main intent direction of the target user is determined by combining the intent intensity descriptor. Based on the main intent direction, integrate relevant user inquiry statements and customer service response statements, extract key information, and summarize the core needs of the target user in the entire historical dialogue. By analyzing the topic association sequence and topic switching frequency in the dialogue topic association features, the occurrence order, duration and switching nodes of different topics in the dialogue can be analyzed. The thematic development knowledge path of the entire historical dialogue is constructed based on the order of appearance of the themes, their duration, and transition points; By summarizing core user needs and integrating knowledge paths developed from dialogue topics, the semantic understanding results of historical dialogues are generated. The user core needs information is hierarchically divided, the core needs are decomposed into multi-level needs, the subordinate relationship between the needs at each level is determined, and the hierarchical division result is obtained. The knowledge path of topic development is divided into stages based on the topic transition nodes, and the entire dialogue process is divided into different stages. Each stage corresponds to one or more closely related topics, resulting in the stage division results. The historical dialogue semantic understanding results are optimized based on the hierarchical division results and the stage division results to form the final historical dialogue semantic understanding results.

2. The dialogue interaction method based on long-history dialogue semantic understanding according to claim 1, characterized in that, The process of performing deep semantic mining on the historical dialogue information set to obtain the dialogue intent features and dialogue topic association features of the target user includes: The user inquiry statements and customer service response statements in the historical dialogue information set are segmented according to the natural pauses and semantic integrity of the statements, and divided into multiple dialogue statement units. Each dialogue statement unit retains the original context position information. Feature words related to e-commerce transactions are extracted from the multiple dialogue statement units, and a core feature word set is determined through word frequency statistics and semantic importance evaluation. The feature words include product names, transaction process terms, and question description words. Semantic analysis of the core feature word set is performed in conjunction with the context to determine the basic intent category corresponding to each user's inquiry statement; Based on the aforementioned basic intent category analysis, the topic continuity between different rounds of dialogue statement units is analyzed, and a topic relevance parameter is generated by calculating the topic similarity parameter. The topic relevance parameter is used to represent the degree of topic relevance between adjacent rounds of dialogue. Based on the basic intent category and the topic relevance parameter, the dialogue intent feature and dialogue topic relevance feature of the target user are constructed. The dialogue intent feature includes an intent category sequence and an intent intensity descriptor. The dialogue topic relevance feature includes topic switching frequency and topic duration parameters. The dialogue intent features are weighted and integrated according to the importance of each round of dialogue in the overall historical dialogue, and a first weight is assigned to each round of dialogue. The higher the first weight, the greater the impact of the round of dialogue on the final dialogue intent features. The dialogue topic association features are calibrated temporally. The weighted and integrated dialogue intent features and the temporally calibrated dialogue topic association features are cross-validated. If the target user dialogue tendencies reflected by the two are consistent, the dialogue intent features and the dialogue topic association features are confirmed to be obtained. If the target user dialogue tendencies reflected by the two are different, the feature extraction parameters are readjusted and deep semantic mining is performed again until the dialogue intent features and the dialogue topic association features are obtained.

3. The dialogue interaction method based on long-history dialogue semantic understanding according to claim 2, characterized in that, The process of extracting e-commerce transaction-related feature words from the multiple dialogue statement units and determining the core feature word set through word frequency statistics and semantic importance evaluation includes: Each dialogue sentence unit is subjected to part-of-speech tagging to identify nouns, verbs, and adjectives, and nouns and verbs that can be used as feature words are selected from them; The selected words are compared with a pre-set e-commerce terminology database to match candidate feature words belonging to product names, transaction process terms, and problem descriptions. The total number of times each candidate feature word appears in the historical dialogue information set is counted, and the distribution frequency of each candidate feature word in each round of dialogue is calculated to obtain the frequency distribution feature. Based on the semantic weight of each candidate feature word in the sentence, its relevance to the topic of the dialogue, and its criticality in the e-commerce transaction scenario, the semantic importance of the candidate feature words is evaluated, and each candidate feature word is assigned a corresponding importance score. A screening threshold is set by combining frequency distribution features and importance scores. Candidate feature words with importance scores higher than the screening threshold are selected to form a core feature word set. Each word in the core feature word set retains its original position mark in the dialogue sentence unit. The core feature word set is deduplicated, and the deduplicated core feature word set is ranked according to the degree of semantic association between the core feature words to determine the association ranking result; The internal structure of the core feature word set is optimized based on the correlation ranking results to form the final core feature word set.

4. The dialogue interaction method based on long-history dialogue semantic understanding according to claim 2, characterized in that, The semantic analysis of the core feature word set, combined with the context, determines the basic intent category corresponding to each user's inquiry statement, including: Construct a mapping table between basic intent categories and core feature words. The mapping table records the mapping relationships between product name feature words corresponding to inquiries about product information, transaction process terminology feature words corresponding to inquiries about transaction status, and problem description vocabulary feature words corresponding to inquiries about post-sales issues. For each user inquiry statement, the core feature words contained in the user inquiry statement are extracted from the core feature word set. Based on the mapping relationship recorded in the association mapping table, the candidate basic intent category that the user inquiry statement may correspond to is initially determined. Based on the mapping logic analysis of the association mapping table, the context of the user's inquiry statement is analyzed, and the core feature words and semantic tendencies involved in the adjacent customer service response statements and user inquiry statements are examined to verify whether the initially determined candidate basic intent categories conform to the mapping relationship. If there are multiple candidate basic intent categories, calculate the matching degree between each candidate and the context based on the association strength between the feature words and intent categories in the association mapping table, and select the candidate basic intent category with the highest matching degree as the basic intent category corresponding to the user's inquiry statement. Arrange the basic intent categories corresponding to each user's inquiry statement in the order of the dialogue rounds to form a basic intent category sequence; Calculate the proportion of each basic intent category in the basic intent category sequence, analyze the distribution characteristics of different intent categories, and the distribution characteristics are positively correlated with the preset association strength of the association mapping table; The basic intent category sequence is adjusted based on the distribution characteristics to form the final basic intent category sequence.

5. The dialogue interaction method based on long-history dialogue semantic understanding according to claim 2, characterized in that, The step of analyzing the topic continuity between different rounds of dialogue statements based on the basic intent category, and generating topic relevance parameters by calculating topic similarity parameters, includes: A topic feature vector is constructed for each basic intent category. The topic feature vector consists of the core feature words corresponding to the basic intent category and their second weights. The second weights are determined based on the importance scores of the feature words. Select dialogue statement units from adjacent rounds and extract the topic feature vectors of the basic intent categories corresponding to the dialogue statement units from adjacent rounds respectively; Calculate the cosine similarity between two topic feature vectors to obtain the topic similarity parameter. The larger the topic similarity parameter, the closer the topic is related between the two rounds of dialogue. The association level is divided according to the magnitude of the topic similarity parameter, and different association levels correspond to different topic association parameter values; Record the topic relevance parameters of adjacent rounds of dialogue in turn order to form a topic relevance sequence. At the same time, calculate the total number of topic transitions in the entire historical dialogue information set and combine them with the total number of dialogue rounds to obtain the topic transition frequency. By setting a fixed-length window, a sliding window analysis is performed on the topic relevance sequence. The topic relevance sequence is traversed, and the average relevance parameter within the window is calculated. The topic relevance sequence is adjusted based on the average relevance parameter, and the final topic relevance parameter is generated by combining the adjusted topic relevance sequence with the topic switching frequency.

6. The dialogue interaction method based on long-history dialogue semantic understanding according to claim 1, characterized in that, The analysis of the topic association sequence and topic transition frequency in the dialogue topic association features reveals the order of appearance, duration, and transition nodes of different topics in the dialogue, including: Traverse the topic relevance sequence and determine consecutive significantly related dialogue segments based on the topic relevance parameter value. Each significantly related dialogue segment corresponds to a continuous topic. Record the start and end rounds of each significantly related dialogue segment in the historical dialogue, calculate the difference between the end round and the start round, and obtain the duration of the ongoing topic; The order in which different topics appear in historical dialogues is determined based on the starting round order of each significantly related dialogue segment, and the topic names are arranged in order. Find locations in the topic relevance sequence where the relevance parameter value suddenly decreases, and use these as potential topic transition nodes. The rationality of the potential conversion nodes is verified by combining the topic conversion frequency. Once the rationality is confirmed, they are marked as official conversion nodes, and the corresponding topics before and after the nodes are determined. Organize the order of appearance, duration, and transition points of different topics into structured data; The duration of each topic is normalized, and the duration is converted into a proportion of the total dialogue duration. The duration parameter in the structured data is then adjusted based on the normalization result.

7. The dialogue interaction method based on long-history dialogue semantic understanding according to claim 1, characterized in that, The step of generating an adapted dialogue interaction strategy based on the semantic understanding results of the historical dialogue includes: The core user needs are summarized from the semantic understanding results of the historical dialogues to determine the information domains covered by the response content and the key information of the response content. Analyze the knowledge path of the dialogue topic, determine the current stage of development and possible extension directions of the dialogue topic, and determine the direction of guidance for the dialogue based on the core needs of the users. By referencing the target user's inquiry frequency, inquiry sentence length, and response speed in historical dialogues, the target user's dialogue rhythm habits are assessed to determine the rhythm control method information to be adopted in new dialogue interactions. The rhythm control method information includes control information on response speed and response sentence length. Based on the key information of the response content, the guidance direction, and the rhythm control method, a corresponding dialogue interaction strategy is constructed. Priorities are set for the key information of the response content according to different user core needs, and the intensity of the guidance direction is adjusted according to different stages of topic development. The dialogue interaction strategy is optimized based on the set priority and intensity adjustment results to form the final dialogue interaction strategy.

8. The dialogue interaction method based on long-history dialogue semantic understanding according to claim 1, characterized in that, The process of executing the dialogue interaction strategy to achieve dialogue interaction with the target user, outputting response statements that conform to the dialogue interaction strategy, and receiving new input statements from the target user includes: Based on the emphasis information of the response content in the dialogue interaction strategy, a matching response template is selected from the preset response template library, and information related to the user's core needs and guiding statements corresponding to the guidance direction are filled in to generate an initial response statement; The initial response statement is adjusted according to the rhythm control information, and the adjusted response statement is output. After outputting the adjusted response statement, monitor the target user's input status in real time, wait to receive new input statements from the target user, and record the content and receiving time of the new input statements from the target user. Add the target user's new input statement to the historical dialogue information set; By analyzing the target user's response speed after receiving the adjusted response statement and the content of the target user's new input statement, the effect of the adjusted response statement is evaluated. If the evaluation result shows that the response statement does not meet the target user's expectations, the response content focus information or rhythm control method information in the dialogue interaction strategy is adjusted, and the response statement is regenerated. The regenerated response is output to the target user, and new input from the target user is received again until the dialogue interaction is completed.

9. A dialogue interaction system based on long-history dialogue semantic understanding, characterized in that, The system includes a processor and a memory, the memory being connected to the processor. The memory is used to store programs, instructions, or code, and the processor is used to execute the programs, instructions, or code in the memory to implement the dialogue interaction method based on long-history dialogue semantic understanding as described in any one of claims 1-8.

Images