Store sales real-time interaction assistance system and method based on AI agent

Abstract

The application relates to the technical field of sales interaction assistance, and particularly discloses a store sales real-time interaction assistance system and method based on an AI intelligent agent, which comprises the following modules: a demand interpretation module that extracts demand-related texts of a store customer from real-time voice of the store customer, interprets the demand-related texts of the customer to obtain customer demand; an initial guide assistance module that generates initial guide language based on customer demand and store inventory information by using a preset AI model; a product interest analysis module that is used for obtaining emotion capture information and strong attention action recognition information of the store customer in the process of receiving the initial guide language, and generates a real-time product interest heat map of the store customer; and an advanced guide assistance module that generates advanced guide language based on the latest demand-related texts of the store customer and the real-time product interest heat map until the guide task is completed; and the interaction quality of store sales is comprehensively improved.

Description

Technical Field

[0001] This invention relates to the field of sales interaction assistance technology, and in particular to a real-time interactive assistance system and method for store sales based on AI intelligent agents. Background Technology

[0002] In today's fiercely competitive retail market, the efficiency and quality of store sales are crucial for the survival and development of businesses. Customer demands are becoming increasingly diverse and personalized, posing numerous challenges to traditional store sales models. Quickly and accurately understanding customer needs and providing targeted product recommendations and high-quality service experiences has become key to improving store sales performance. In this context, a real-time interactive auxiliary system and method for store sales based on AI agents has emerged and is of great significance. With the rapid development of artificial intelligence technology, AI agents have demonstrated powerful capabilities in natural language processing, image recognition, and other areas. Applying them to store sales scenarios aligns with the trend of digital transformation in the retail industry and is expected to become an important tool for enhancing the competitiveness of stores in the future, with broad application prospects in the retail sector.

[0003] Existing in-store sales support technologies struggle to deeply understand customer needs, leading to inaccurate comprehension of customer intentions. In the sales assistant support stage, the inability to effectively utilize real-time inventory information from the CRM system to generate appropriate sales scripts results in a lack of targeted recommendations. Furthermore, insufficient attention is paid to capturing customer feedback during sales script delivery, hindering the comprehensive understanding of real-time changes in customer needs and interests, and preventing the generation of effective, advanced sales scripts. This makes it difficult to meet constantly evolving customer demands, impacting sales efficiency and customer experience.

[0004] Therefore, this invention proposes a real-time interactive assistance system and method for store sales based on AI intelligent agents. Summary of the Invention

[0005] This invention provides a real-time interactive assistance system and method for in-store sales based on AI intelligent agents. It can analyze customers' voice, actions, and emotions in real time to deeply understand their needs. By integrating with a CRM system to obtain inventory information, it uses AI models to provide customers with accurate and relevant product recommendations. Simultaneously, it continuously tracks customer feedback during the sales process, dynamically adjusts recommendation strategies, and provides advanced sales assistance, effectively improving customer satisfaction and purchase conversion rates.

[0006] This invention provides a real-time interactive assistance system for store sales based on AI intelligent agents, comprising:

[0007] The demand interpretation module is used to extract the text related to the needs of store customers from their real-time voice recordings, and to interpret the text related to the needs of customers to obtain their needs.

[0008] The initial sales guide assistance module is used to connect to the CRM system to read store inventory information and use a preset AI model to generate initial sales guide scripts based on customer needs and store inventory information.

[0009] The product interest analysis module is used to acquire emotional capture information and strong attention action recognition information of store customers during the process of receiving the initial sales guide script, and to generate a real-time product interest heatmap of store customers based on the emotional capture information and strong attention action recognition information.

[0010] The advanced sales guide assistance module is used to generate advanced sales guide scripts based on the latest needs of store customers, including text and real-time product interest heatmaps, until the sales guide task is completed.

[0011] The preferred requirement interpretation module includes:

[0012] The sentence segmentation and word segmentation processing submodule is used to convert the real-time voice of store customers into voice text, and to perform sentence segmentation and word segmentation processing on the voice text to obtain all customer input sentences and the word set of each customer input sentence;

[0013] The vertical correlation determination submodule is used to determine the correlation between each word in each word set and the store's products and services, which is used as the vertical correlation of each word in each word set;

[0014] The horizontal correlation determination submodule is used to obtain the horizontal correlation of each word in each word set based on the correlation between each word in each word set and all remaining words in all word sets except the current word;

[0015] The requirement involves a two-dimensional submodule, which uses the vertical and horizontal correlation of each word as the vertical and horizontal coordinate values ​​of the corresponding word in a preset two-dimensional coordinate system, respectively, to generate a marker point for each word in the preset coordinate system, and to generate the required two-dimensional graph based on the marker points of all words in the preset coordinate system.

[0016] The requirement-related text extraction submodule is used to extract the requirement-related text of store customers from the voice text based on the requirement-related two-dimensional diagram.

[0017] The requirement involves a text interpretation submodule, which is used to interpret the text related to customer requirements to obtain customer requirements.

[0018] Preferably, the horizontal correlation determination submodule includes:

[0019] The order-based classification unit is used to treat the correlation between each word in each word set and each word in the corresponding word set other than the current word as the same-order correlation between the two words. At the same time, the correlation between each word in each word set and each word in all word sets except the current word set is treated as the different-order correlation between the two words.

[0020] The tree structure building unit is used to filter out all strongly related words from all words based on the vertical correlation of all words, and build a strong related word development tree structure based on the occurrence sequence and original association relationship of all strongly related words;

[0021] The semantic aggregation unit is used to treat all strongly related words as cluster centers and to cluster all words in all word sets based on the degree of association between each pair of words in all word sets, so as to obtain the semantic aggregation word set of each strongly related word.

[0022] The horizontal correlation determination unit is used to determine the horizontal correlation of each word in each word set based on the strong correlation word development tree structure, the semantic aggregation word set of all strong correlation words, and the correlation between pairs of words in all word sets.

[0023] Preferably, the horizontal correlation determination unit includes:

[0024] The first reinforcement calculation subunit is used to determine the reinforcement factor of the same-order correlation between two words based on the correlation between the two words and the strongly related words of the corresponding semantic aggregation word set when two words in all word sets belong to the same word set and the same semantic aggregation word set. Based on the same-order correlation between the two words and the corresponding reinforcement factor, the reinforcement correlation between the two words is determined.

[0025] The second reinforcement calculation subunit is used to determine the reinforcement factor of the same-order correlation between two words when two words in all word sets belong to the same word set but not to the same semantic aggregation word set. This is based on the correlation between the two words and the strong related words of their respective semantic aggregation word sets, the correlation between the strong related words of their respective semantic aggregation word sets, and the cross-set factor of the strong related words of their respective semantic aggregation word sets in the strong related word development tree structure. Based on the same-order correlation between the two words and the corresponding reinforcement factor, the reinforcement correlation between the two words is determined.

[0026] The third reinforcement calculation subunit is used to determine the reinforcement factor of the heterogeneous correlation between two words when two words in all word sets do not belong to the same word set and do not belong to the same semantic aggregation word set. This is based on the correlation between the two words and the strong related words in their respective semantic aggregation word sets, the correlation between the strong related words in their respective semantic aggregation word sets, and the cross-layer factor of the strong related words in the strong related word development tree structure. Based on the heterogeneous correlation between the two words and the corresponding reinforcement factor, the reinforcement correlation between the two words is determined.

[0027] The horizontal correlation determination subunit is used to take the average of the enhanced correlation between each word in each word set and all remaining words in all word sets except the current word as the horizontal correlation of each word in each word set.

[0028] Preferably, the requirements involve a text extraction submodule, including:

[0029] The segmentation span value determination unit is used to determine the distance between the marker point closest to the origin and the marker point farthest from the origin in the two-dimensional diagram involving the requirements, and use it as the span value of the two-dimensional diagram involving the requirements. The product of the span value of the two-dimensional diagram involving the requirements and the preset ratio, and the sum of the distances between the marker point closest to the origin and the origin, are used as the segmentation span value of the two-dimensional diagram involving the requirements.

[0030] The requirement involves a marker point filtering unit, which is used to divide a circular area in the preset coordinate system with the origin as the origin and the span value as the radius, and to treat all marker points in the two-dimensional graph of the requirement that are outside the circular area as all marker points of the requirement.

[0031] The demand-related text extraction unit is used to extract the demand-related text of store customers from the speech text based on all demand-related markers in the demand-related 2D graph.

[0032] Preferably, the requirements involve a text extraction unit, including:

[0033] The cluster location marker subunit is used to determine the distribution of marker points corresponding to all words in the same word set in a preset two-dimensional coordinate system, which serves as the cluster location of each word set;

[0034] The deduplication requirement judgment subunit is used to calculate the similarity between the cluster positions of every two word sets, and merge all customer input statements corresponding to all word sets whose similarity exceeds the similarity threshold as the text to be deduplicated and refined, and treat each customer input statement that is not merged as the text to be refined.

[0035] The deduplication and refinement sub-unit is used to deduplicatize and refine the text to be deduplicated and refined, and at the same time, refine the text to be refined to obtain the text related to the needs of store customers.

[0036] Preferably, the product interest analysis module includes:

[0037] The information capture submodule is used to read emotional capture information and strong attention action recognition information of store customers in the process of receiving the initial sales guide's speech from the in-store surveillance video.

[0038] The Interested Product Distribution Map Generation Submodule is used to determine the sequence of products and services of interest to users based on strong attention action recognition information, and generate a network distribution map of the user's interested products by combining the correlation between products and services in the store.

[0039] The product interest heatmap generation submodule is used to assign bidirectional interest values ​​to the network distribution map of users' interested products based on emotion capture information, thereby obtaining a real-time product interest heatmap of store customers.

[0040] Preferably, the product interest heatmap generation submodule includes:

[0041] The emotion category recognition unit is used to generate an emotion category sequence of store customers during the initial sales guide dialogue based on emotion capture information, and to perform bidirectional intensity evaluation on each emotion in the emotion category sequence to obtain the emotion category and intensity sequence.

[0042] The bidirectional intensity determination unit is used to align the emotion category and intensity sequence with the user's sequence of products and services of interest in time, and determine all the emotion categories and corresponding bidirectional intensities involved in each product and service of interest to the user;

[0043] The bidirectional assignment tagging unit is used to generate bidirectional interest assignment results for each product service of interest based on all the emotion categories involved and the corresponding bidirectional intensity. Based on the bidirectional interest assignment results of all product services of interest, the network distribution map of the user's products of interest is marked to obtain a real-time product interest heatmap of store customers.

[0044] Preferred advanced shopping guide assistance modules include:

[0045] The thermal level zone molecular module is used to classify the real-time product interest heatmap into thermal levels and obtain the local real-time product interest local thermal map under each thermal level.

[0046] The advanced sales guide assistance submodule is used to input the local real-time product interest heatmaps under all heat levels and the latest needs of store customers into the preset AI model to generate advanced sales guide scripts until the sales guide task is completed.

[0047] This invention provides a real-time interactive assistance method for store sales based on an AI agent, comprising:

[0048] S1: Extract the text related to customer needs from the real-time voice of customers in the store, and interpret the text related to customer needs to obtain customer needs.

[0049] S2: Connect to the CRM system to read store inventory information and use a preset AI model to generate initial sales scripts based on customer needs and store inventory information;

[0050] S3: Acquire emotional capture information and strong attention action recognition information of store customers during the process of receiving the initial sales guide script, and generate a real-time product interest heatmap of store customers based on the emotional capture information and strong attention action recognition information.

[0051] S4: Based on the latest needs of store customers, generate advanced sales scripts involving text and real-time product interest heatmaps until the sales task is completed.

[0052] The beneficial effects of this invention compared to existing technologies are as follows: The demand interpretation module extracts and interprets customer needs from real-time voice, accurately grasping customer intentions and providing accurate direction for subsequent sales guidance, avoiding communication deviations caused by misunderstandings of needs, and improving the customer shopping experience. The initial sales guidance assistance module connects to the CRM system to obtain inventory information and combines it with a preset AI model to generate initial sales scripts. This ensures that sales recommendations are based on actual inventory, avoiding the embarrassment of recommending out-of-stock items, and utilizes the AI ​​model to make the scripts more targeted and professional, improving sales efficiency. The product interest analysis module generates real-time product interest heatmaps by capturing customer emotions and strong attentional actions, which can deeply understand customers' interests in products and grasp changes in customer psychology in real time, providing data support for further optimization of sales guidance. The advanced sales guidance assistance module generates advanced sales scripts based on the latest demand text and interest heatmaps, continuously adapting to customers' dynamic needs and interests, making the sales process more personalized and precise.

[0053] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in this application.

[0054] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0055] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0056] Figure 1 This is a schematic diagram of a real-time interactive assistance system for store sales based on an AI intelligent agent, as described in an embodiment of the present invention.

[0057] Figure 2 This is a schematic diagram of the requirements interpretation module in an embodiment of the present invention;

[0058] Figure 3 This is a schematic diagram of the product interest analysis module in an embodiment of the present invention;

[0059] Figure 4This is a schematic diagram of the advanced shopping guide auxiliary module in an embodiment of the present invention. Detailed Implementation

[0060] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0061] Example 1: This invention provides a real-time interactive assistance system for store sales based on AI intelligent agents, referencing... Figure 1 ,include:

[0062] The demand interpretation module is used to extract the text related to the needs of store customers from their real-time voice recordings, and to interpret the text related to the needs of customers to obtain their needs.

[0063] The initial sales guide assistance module is used to connect to the CRM system to read store inventory information and use a preset AI model to generate initial sales guide scripts based on customer needs and store inventory information.

[0064] The product interest analysis module is used to acquire emotional capture information and strong attention action recognition information of store customers during the process of receiving the initial sales guide script, and to generate a real-time product interest heatmap of store customers based on the emotional capture information and strong attention action recognition information.

[0065] The advanced sales guide assistance module is used to generate advanced sales guide scripts based on the latest needs of store customers, including text and real-time product interest heatmaps, until the sales guide task is completed.

[0066] In this embodiment, the demand-related text refers to the demand-related text extracted from the real-time voice of customers in the store, such as the description of the product mentioned by the customer.

[0067] In this embodiment, the present invention relates to the collection and processing of personal privacy information (including but not limited to customer voice data). In order to comply with relevant laws and regulations, the collection of the above information must strictly follow the informed consent principle:

[0068] Before collecting information, the implementation device of this invention must inform the information subject of the purpose of collection, the type of information to be collected, the scope of use, the storage period, and the information subject's rights to withdraw consent, query, and delete in a clear and perceptible manner (such as voice prompts or visual interface display);

[0069] The device will only initiate the privacy information collection process after the data subject has explicitly expressed their consent through active confirmation (such as clicking the "agree" button or verbally agreeing).

[0070] The data subject has the right to withdraw consent at any time through the device's preset operation path. After consent is withdrawn, the device will immediately terminate the data collection and delete or anonymize the collected privacy information to ensure that the data subject's rights are not infringed.

[0071] In this embodiment, customer needs refer to the clear expectations and demands of customers for products or services after interpreting the text related to the needs.

[0072] In this embodiment, the CRM system, or Customer Relationship Management System, can provide information such as store inventory to assist sales activities.

[0073] In this embodiment, store inventory information refers to the storage status data such as the quantity and location of products in the store.

[0074] In this embodiment, the preset AI model is an artificial intelligence model that has been trained to generate sales pitches based on customer needs (and the text related to those needs) and inventory information (or local real-time product interest heatmaps under all heat levels); for example, an NLP model based on deep learning.

[0075] It is trained using customer demand texts and current inventory information (or local real-time product interest heatmaps under all heat levels) from a large number of successful sales cases as input samples and corresponding sales scripts as output samples.

[0076] In this embodiment, the initial sales pitch is a preliminary sales guide language generated by a preset AI model combining customer needs and inventory information.

[0077] In this embodiment, the emotion capture information is obtained from surveillance video, specifically regarding the customer's emotional state during the sales pitch.

[0078] In this embodiment, the strong attention action recognition information is identified from surveillance video, reflecting the customer's strong attention to the product. Examples include the duration of gaze and lingering in front of the product, or the action of picking up an item.

[0079] In this embodiment, the real-time product interest heatmap is a visualization generated based on emotion and strong attention action recognition information, showing the customer's real-time interest in the product.

[0080] In this embodiment, the advanced sales guide script is the subsequent sales guide language generated based on the latest demand text and real-time product interest heatmap.

[0081] In this embodiment, the latest demand refers to the text related to the demand in the new statement of the store customer.

[0082] In this embodiment, the end of the sales guide task refers to the customer making a decision such as purchasing or abandoning the purchase, and the sales guide process is completed.

[0083] Example 2: Based on Example 1, the requirements interpretation module, refer to... Figure 2 ,include:

[0084] The sentence segmentation and word segmentation processing submodule is used to convert the real-time voice of store customers into voice text, and to perform sentence segmentation and word segmentation processing on the voice text to obtain all customer input sentences and the word set of each customer input sentence;

[0085] The vertical correlation determination submodule is used to determine the correlation between each word in each word set and the store's products and services, which is used as the vertical correlation of each word in each word set;

[0086] The horizontal correlation determination submodule is used to obtain the horizontal correlation of each word in each word set based on the correlation between each word in each word set and all remaining words in all word sets except the current word;

[0087] The requirement involves a two-dimensional submodule, which uses the vertical and horizontal correlation of each word as the vertical and horizontal coordinate values ​​of the corresponding word in a preset two-dimensional coordinate system, respectively, to generate a marker point for each word in the preset coordinate system, and to generate the required two-dimensional graph based on the marker points of all words in the preset coordinate system.

[0088] The requirement-related text extraction submodule is used to extract the requirement-related text of store customers from the voice text based on the requirement-related two-dimensional diagram.

[0089] The requirement involves a text interpretation submodule, which is used to interpret the text related to customer requirements to obtain customer requirements.

[0090] In this embodiment, customer input statements refer to individual statements obtained by converting the real-time speech of customers in the store into speech-to-text and then processing it into sentences. For example, if a customer says, "I want to buy a pair of running shoes, preferably lightweight and breathable," after sentence processing, "I want to buy a pair of running shoes" and "preferably lightweight and breathable" are considered customer input statements.

[0091] In this embodiment, determining the correlation between each word in each word set and the store's products and services refers to analyzing the closeness of the connection between each word in each word set and the products or services offered by the store. For example, if the word set contains words such as "sports shoes," "lightweight," and "running," "sports shoes" is directly related to the products sold in the store, so the correlation may be high; "lightweight" describes a product attribute and is also related to the product service; "running" is related to the product's usage scenario and is also related to the store's products and services. The specific correlation value needs to be determined through a specific method. For example, the average correlation between each word and all words in the preset word group corresponding to the store's products and services (obtained by statistically analyzing the co-occurrence frequency of two words in a large amount of text data) can be used as the correlation between the word and the store's products and services.

[0092] In this embodiment, the correlation between each word in each word set and all remaining words in all word sets except the current word is obtained by statistically analyzing the co-occurrence frequency of two words in a large amount of text data. This reflects the tightness of semantic and logical connections between a word and all other words (within the same word set and different word sets). For example, "sports shoes" and "lightweight" may be semantically related, and "sports backpack" may have a logical association of being used together. By calculating this correlation, a more comprehensive understanding of the relationships between words can be obtained, and potential connections in the expression of customer needs can be uncovered.

[0093] The beneficial effects of the above technologies are as follows: The sentence and word segmentation processing submodule converts real-time speech into text and processes it, breaking down complex speech information into easily analyzable units, laying the foundation for subsequent accurate analysis. The vertical correlation determination submodule clarifies the correlation between each word and the store's products and services, helping to understand the key points of customer speech from a product and service perspective and grasp the direction of potential customer needs. The horizontal correlation determination submodule considers the correlation between words and other words, uncovering hidden relationships between words, resulting in a more comprehensive understanding of customer needs. The demand-related two-dimensional visualization submodule presents the vertical and horizontal correlations in the form of a two-dimensional graph, visualizing abstract relationships and facilitating an intuitive grasp of the overall demand. The demand-related text extraction submodule accurately extracts relevant text based on the two-dimensional graph, avoiding interference from irrelevant information and improving the accuracy of demand extraction. The demand-related text interpretation submodule interprets the extracted text to derive customer needs, making the entire process logically coherent and ultimately accurately obtaining customer needs, providing a more accurate basis for subsequent sales guidance, and effectively improving the intelligence and practicality of the store's real-time interactive auxiliary system.

[0094] Example 3: Based on Example 2, the horizontal correlation degree is determined by the sub-module, referring to... Figure 2 ,include:

[0095] The order-based classification unit is used to treat the correlation between each word in each word set and each word in the corresponding word set other than the current word as the same-order correlation between the two words. At the same time, the correlation between each word in each word set and each word in all word sets except the current word set is treated as the different-order correlation between the two words.

[0096] The tree structure building unit is used to filter out all strongly related words from all words based on the vertical correlation of all words, and build a strong related word development tree structure based on the occurrence sequence and original association relationship of all strongly related words;

[0097] The semantic aggregation unit is used to treat all strongly related words as cluster centers and to cluster all words in all word sets based on the degree of association between each pair of words in all word sets, so as to obtain the semantic aggregation word set of each strongly related word.

[0098] The horizontal correlation determination unit is used to determine the horizontal correlation of each word in each word set based on the strong correlation word development tree structure, the semantic aggregation word set of all strong correlation words, and the correlation between pairs of words in all word sets.

[0099] In this embodiment, based on the vertical correlation of all words, all strongly related words are selected from all words. This involves setting a standard (threshold) based on the degree of association between each word and the store's products and services (i.e., vertical correlation). Words that are closely related to the store's products and services and meet or exceed this standard (i.e., the correlation is not less than the threshold) are selected. These selected words are the strongly related words. For example, among words describing customer needs, words directly related to the products sold in the store, such as "mobile phone" and "computer," will be identified as strongly related words if their vertical correlation is higher than the set value.

[0100] In this embodiment, a strong-association word development tree structure is constructed based on the occurrence sequence and original association relationships of all strong-associations. This tree structure is built according to the order in which strong-associations appear in customer statements (occurrence sequence) and their inherent semantic, logical, and other connections (original association relationships). In this structure, strong-associations act as nodes, and the relationships between them are represented by lines, thus demonstrating the development path and interrelationships between strong-associations. For example, if "computer" appears first, followed by "computer accessories," then in the tree structure, "computer" might be the parent node, and "computer accessories" might be the child node connected to it, reflecting the association and sequence of events between the two.

[0101] In this embodiment, all strongly related words are used as cluster centers, and all words in all word sets are clustered based on the correlation between pairs of words in each word set to obtain a semantic aggregation word set for each strongly related word. That is, the previously selected strongly related words are used as core points. Then, based on the semantic and logical correlation between different words in each word set, all words are divided into sets centered on different strongly related words. For example, using "computer" as a strongly related word, words highly associated with "computer," such as "monitor," "keyboard," and "CPU," will be grouped into a semantic aggregation word set centered on "computer," forming a set of semantically similar or related words. The specific implementation process includes:

[0102] All strongly related words are used as cluster centers, and all remaining words in all word sets except the strongly related words are randomly assigned to each strongly related word to obtain a cluster containing a single strongly related word and the rest of the words;

[0103] The mean of the correlation between all pairs of words in each cluster is taken as the cluster degree of each cluster;

[0104] Determine whether the average clustering degree of all clusters is not less than a preset threshold. If so, treat all words in each cluster as the semantic aggregate word set of the corresponding strongly related words. Otherwise, arbitrarily group all words in all word sets except the strongly related words with each strongly related word to obtain a new cluster containing a single strongly related word and the remaining words. Continue until the clustering degree of all new clusters is not less than the preset threshold. Then, treat all words in each new cluster as the semantic aggregate word set of the corresponding strongly related words.

[0105] The beneficial effects of the above technologies are as follows: The hierarchical classification unit clearly distinguishes between the same-level and different-level correlations, comprehensively considering the correlations between words at different levels, providing a more detailed dimension for accurate analysis of inter-word relationships, and avoiding the limitations of single-dimensional analysis. The tree structure construction unit, by selecting strongly correlated words and constructing a development tree structure, sorts out key information from the perspective of time sequence and original correlation, forming a clear logical framework, which helps to grasp the deep-seated developmental relationships between words and provides an important basis for understanding the evolution of customer needs. The semantic aggregation unit clusters words with strongly correlated words as cluster centers, effectively integrating semantically similar words, aggregating scattered words according to semantics, making semantic expression clearer, and facilitating the understanding of customer needs from an overall semantic level. The horizontal correlation determination unit comprehensively integrates the development tree structure of strongly correlated words, the semantic aggregation word set, and the inter-word correlations to determine the horizontal correlation. This comprehensive integration of multi-dimensional information greatly improves the accuracy of determining the horizontal correlation of each word, thereby making the interpretation of needs more in-depth and accurate, and enabling the store sales real-time interactive auxiliary system to better understand customer needs.

[0106] Example 4: Based on Example 3, the horizontal correlation degree determination unit, refer to Figure 2 ,include:

[0107] The first reinforcement calculation subunit is used to determine the reinforcement factor of the same-order correlation between two words based on the correlation between the two words and the strongly related words of the corresponding semantic aggregation word set when two words in all word sets belong to the same word set and the same semantic aggregation word set. Based on the same-order correlation between the two words and the corresponding reinforcement factor, the reinforcement correlation between the two words is determined.

[0108] The second reinforcement calculation subunit is used to determine the reinforcement factor of the same-order correlation between two words when two words in all word sets belong to the same word set but not to the same semantic aggregation word set. This is based on the correlation between the two words and the strong related words of their respective semantic aggregation word sets, the correlation between the strong related words of their respective semantic aggregation word sets, and the cross-set factor of the strong related words of their respective semantic aggregation word sets in the strong related word development tree structure. Based on the same-order correlation between the two words and the corresponding reinforcement factor, the reinforcement correlation between the two words is determined.

[0109] The third reinforcement calculation subunit is used to determine the reinforcement factor of the heterogeneous correlation between two words when two words in all word sets do not belong to the same word set and do not belong to the same semantic aggregation word set. This is based on the correlation between the two words and the strong related words in their respective semantic aggregation word sets, the correlation between the strong related words in their respective semantic aggregation word sets, and the cross-layer factor of the strong related words in the strong related word development tree structure. Based on the heterogeneous correlation between the two words and the corresponding reinforcement factor, the reinforcement correlation between the two words is determined.

[0110] The horizontal correlation determination subunit is used to take the average of the enhanced correlation between each word in each word set and all remaining words in all word sets except the current word as the horizontal correlation of each word in each word set.

[0111] In this embodiment, the strengthening factor of the same-order correlation between two corresponding words is determined based on the correlation between the two corresponding words and the strongly related words in the corresponding semantic aggregate word set. That is, the average of the correlation between the two corresponding words and the strongly related words in the corresponding semantic aggregate word set is used as the strengthening factor of the same-order correlation between the two corresponding words.

[0112] In this embodiment, the reinforcement correlation degree between two words is determined based on the same (different) order correlation degree and the corresponding reinforcement factor. This means that the same or different order correlation degree of the two words is used to perform a calculation (such as multiplication) with the corresponding reinforcement factor to obtain an adjusted correlation degree. This new correlation degree is the reinforcement correlation degree, which more accurately reflects the actual degree of closeness of the two words after considering the reinforcement factor.

[0113] In this embodiment, the strengthening factor of the same-order correlation between two corresponding words is determined based on the correlation between the two corresponding words and the strongly related words in their respective semantic aggregation word sets, the correlation between the two corresponding words and the strongly related words in their respective semantic aggregation word sets, and the cross-set factor of the strongly related words in the strong related word development tree structure of the semantic aggregation word sets of the two corresponding words. This is done by multiplying the mean of the correlation between the two corresponding words and the strongly related words in their respective semantic aggregation word sets, and the correlation between the two corresponding words and the strongly related words in their respective semantic aggregation word sets, by 1 and the difference between 1 and the cross-set factor of the strongly related words in the strong related word development tree structure of the semantic aggregation word sets of the two corresponding words, and using this result as the strengthening factor of the same-order correlation between the two corresponding words.

[0114] The cross-set factor is the difference between the number of edges in the shortest path of the two words in the strong association word development tree structure and the total number of edges in the strong association word development tree structure.

[0115] In this embodiment, the strengthening factor of hetero-order correlation between two words is determined based on the correlation between each of the two words and the strong related words in their respective semantic aggregation word sets, the correlation between the strong related words in their respective semantic aggregation word sets, and the cross-level factor of the strong related words in the strong related word development tree structure of the semantic aggregation word sets of the two words. This is done by multiplying the mean of the correlation between the two words and the strong related words in their respective semantic aggregation word sets and the correlation between the strong related words in their respective semantic aggregation word sets by 1 and the difference between 1 and the cross-level factor of the strong related words in the strong related word development tree structure of the semantic aggregation word sets of the two words. This result is used as the strengthening factor of same-order correlation between the two words.

[0116] The cross-layer factor is the quotient of the number of layers that the two words cross in the strong association word development tree structure and the total number of layers in the strong association word development tree structure.

[0117] The beneficial effects of the above technologies are as follows: For words within the same word set and semantic aggregation set, the first reinforcement calculation subunit determines the reinforcement factor and correlation degree by leveraging the correlation with strongly related words, accurately uncovering the close connections between words within the same semantic category. For words within the same word set but different semantic aggregation sets, the second reinforcement calculation subunit determines the reinforcement correlation degree by comprehensively considering multiple factors, fully assessing the potential connections between different semantic modules. For words within different word sets and different semantic aggregation sets, the third reinforcement calculation subunit determines the reinforcement correlation degree by combining multiple factors, effectively capturing long-distance semantic connections. Finally, the horizontal correlation degree determination subunit uses the average reinforcement correlation degree as the horizontal correlation degree, comprehensively integrating various types of word association information. This meticulous and comprehensive calculation method enables the system to more accurately grasp the relative importance and correlation of words in customer demand expressions, helping the demand interpretation module accurately understand customer needs and providing strong support for the sales guidance process.

[0118] Example 5: Based on Example 2, the requirements involve a text extraction submodule, see reference. Figure 2 ,include:

[0119] The segmentation span value determination unit is used to determine the distance between the marker point closest to the origin and the marker point farthest from the origin in the two-dimensional diagram involving the requirements, and use it as the span value of the two-dimensional diagram involving the requirements. The product of the span value of the two-dimensional diagram involving the requirements and the preset ratio, and the sum of the distances between the marker point closest to the origin and the origin, are used as the segmentation span value of the two-dimensional diagram involving the requirements.

[0120] The requirement involves a marker point filtering unit, which is used to divide a circular area in the preset coordinate system with the origin as the origin and the span value as the radius, and to treat all marker points in the two-dimensional graph of the requirement that are outside the circular area as all marker points of the requirement.

[0121] The demand-related text extraction unit is used to extract the demand-related text of store customers from the speech text based on all demand-related markers in the demand-related 2D graph.

[0122] In this embodiment, the preset ratio is a pre-set fixed value, such as 30%.

[0123] The beneficial effects of the above technologies are as follows: The span determination unit calculates the distances to the nearest and farthest markers from the origin in the two-dimensional demand map and combines this with a preset ratio to derive the span value. This method comprehensively considers the distribution range of demand in the two-dimensional map and preset standards, providing a reasonable quantitative basis for filtering demand-related information. The demand-related marker filtering unit divides the area into circular regions using the origin and the span value, and uses the markers outside the circle as demand-related markers. This cleverly categorizes and filters the information in the two-dimensional map, excluding points within the circle that may have weak relevance to demand, focusing on key information, and making the filtered markers more representative of the customer's core needs. The demand-related text extraction unit extracts demand-related text from the audio text based on these filtered markers, effectively avoiding interference from irrelevant information and greatly improving the accuracy of demand-related text extraction. This series of operations enables the system to more accurately grasp customer needs, providing a reliable foundation for subsequent sales script generation and enhancing the practicality and effectiveness of the store's real-time interactive sales support system.

[0124] Example 6: Based on Example 5, the requirements involve a text extraction unit, see reference. Figure 2 ,include:

[0125] The cluster location marker subunit is used to determine the distribution of marker points corresponding to all words in the same word set in a preset two-dimensional coordinate system, which serves as the cluster location of each word set;

[0126] The deduplication requirement judgment subunit is used to calculate the similarity between the cluster positions of every two word sets, and merge all customer input statements corresponding to all word sets whose similarity exceeds the similarity threshold as the text to be deduplicated and refined, and treat each customer input statement that is not merged as the text to be refined.

[0127] The deduplication and refinement sub-unit is used to deduplicatize and refine the text to be deduplicated and refined, and at the same time, refine the text to be refined to obtain the text related to the needs of store customers.

[0128] In this embodiment, calculating the similarity between the cluster positions of every two word sets refers to measuring the similarity between the distribution positions of corresponding marker points of all words in the same word set in a preset two-dimensional coordinate system using a specific algorithm. For example, methods such as spatial distance measurement (i.e., taking the quotient of the average of the coordinate differences between the marker points of two words belonging to two different word sets and a preset length value as the deviation, and taking the difference between 1 and the deviation as the similarity) and feature matching are used to quantify the similarity of the marker point distribution of the two word sets and determine their similarity in expressing customer needs.

[0129] In this embodiment, the similarity threshold is a pre-set numerical standard, such as 0.8.

[0130] In this embodiment, the text to be deduplicated and refined is deduplicated, and the text to be refined is also refined to obtain the text related to the needs of store customers. For the merged text to be deduplicated and refined, duplicate expressions are removed and key information is extracted; key information is also extracted from the unmerged text to be refined. For example, redundant words are removed and core needs descriptions are retained, ultimately resulting in text content that accurately reflects the needs of store customers, providing accurate basis for subsequent sales guidance.

[0131] The beneficial effects of the above technologies are as follows: The cluster location marking subunit determines the cluster location of each word set in a preset two-dimensional coordinate system, identifying word sets through spatial distribution, providing an intuitive and unique perspective for subsequent analysis, and facilitating the understanding of the relationships between word sets from an overall layout perspective. The deduplication requirement judgment subunit calculates the similarity of word set cluster locations, merging customer input statements corresponding to word sets with similarity exceeding a threshold into text to be deduplicated and refined, while those not merged are used as text to be refined. This method effectively identifies statements expressing similar needs, avoids redundant processing, improves information processing efficiency, and also distinguishes unique requirement expressions. The deduplication and refinement subunit performs deduplication and refinement on the two types of text separately, accurately extracting key information, removing redundant content, and further optimizing the extraction quality of text related to requirements. Through these steps, the system can more clearly and accurately sort out customer needs, providing sales guides with more refined and effective information, and improving the relevance and effectiveness of sales scripts.

[0132] Example 7: Based on Example 1, the product interest analysis module, refer to... Figure 3 ,include:

[0133] The information capture submodule is used to read emotional capture information and strong attention action recognition information of store customers in the process of receiving the initial sales guide's speech from the in-store surveillance video.

[0134] The Interested Product Distribution Map Generation Submodule is used to determine the sequence of products and services of interest to users based on strong attention action recognition information, and generate a network distribution map of the user's interested products by combining the correlation between products and services in the store.

[0135] The product interest heatmap generation submodule is used to assign bidirectional interest values ​​to the network distribution map of users' interested products based on emotion capture information, thereby obtaining a real-time product interest heatmap of store customers.

[0136] In this embodiment, reading emotional information and identifying strong attentional actions of customers during the initial sales pitch from in-store surveillance video refers to using video footage captured by surveillance equipment installed in the store, and employing image recognition and analysis technology to extract information related to the emotional state of customers while listening to the salesperson's initial introduction of the product or service, as well as action information that reflects the customer's high level of attention to the product or service. For example, recognizing facial expressions to determine whether the customer is interested, confused, or dissatisfied, while also identifying strong attentional actions such as prolonged staring or reaching out to touch the product.

[0137] In this embodiment, a sequence of products and services of interest to the user is determined based on strong attention action recognition information, and a network distribution map of the user's interested products is generated by combining the correlation between products and services in the store. Specifically, based on strong attention actions identified from surveillance video, the order in which customers focus on products or services is determined, forming a sequence of interested products and services. Then, based on the inherent connections (i.e., correlation) in function, purpose, etc., among various products or services in the store, the relationships between the customer's interested products are graphically displayed, generating a network distribution map. For example, if a customer first focuses on a mobile phone and then on a phone case, the mobile phone and phone case would be treated as nodes in the map, with lines representing the relationship between them formed due to the customer's attention behavior and their inherent association.

[0138] The beneficial effects of the above technologies are as follows: The information capture submodule obtains customer emotion capture information and strong attention action recognition information from store surveillance videos, comprehensively collecting customer reaction data during the initial sales pitch in a non-intrusive manner, providing rich material for a deeper understanding of customer interests. The product interest distribution map generation submodule determines the sequence of products and services that customers are interested in based on the strong attention action recognition information, and generates a network distribution map by combining the product and service correlation, clearly showing the relationship between products that customers are interested in, helping to grasp the overall context of customer interests and uncover potential interest extensions. The product interest heatmap generation submodule assigns bidirectional interest values ​​to the distribution map based on emotion capture information, combining customer emotions with products of interest. The generated real-time product interest heatmap can intuitively reflect the degree of customer interest and emotional tendency for different products, enabling sales personnel to quickly and accurately grasp the current focus of customer interest. This series of operations makes the system's analysis of customer product interests more in-depth and intuitive, providing a strong basis for the generation of advanced sales pitches, helping sales personnel to provide services that are more tailored to customer interests, improving sales interaction effects, and increasing the probability of sales success.

[0139] Example 8: Based on Example 7, the product interest heatmap generation submodule, refer to... Figure 3 ,include:

[0140] The emotion category recognition unit is used to generate an emotion category sequence of store customers during the initial sales guide dialogue based on emotion capture information, and to perform bidirectional intensity evaluation on each emotion in the emotion category sequence to obtain the emotion category and intensity sequence.

[0141] The bidirectional intensity determination unit is used to align the emotion category and intensity sequence with the user's sequence of products and services of interest in time, and determine all the emotion categories and corresponding bidirectional intensities involved in each product and service of interest to the user;

[0142] The bidirectional assignment tagging unit is used to generate bidirectional interest assignment results for each product service of interest based on all the emotion categories involved and the corresponding bidirectional intensity. Based on the bidirectional interest assignment results of all product services of interest, the network distribution map of the user's products of interest is marked to obtain a real-time product interest heatmap of store customers.

[0143] In this embodiment, generating a sequence of emotional categories for customers during the initial sales pitch based on emotion capture information involves using emotion capture information obtained from surveillance video and employing technologies such as facial expression recognition to determine the emotional categories exhibited by customers at different points in time during the initial sales pitch. These emotional categories are then arranged in chronological order to form a sequence. For example, it might be identified that a customer first expresses curiosity, then hesitation, and finally satisfaction; this sequence of "curiosity-hesitation-satisfaction" constitutes the emotional category sequence.

[0144] In this embodiment, the sequence of emotion categories and intensities is temporally aligned with the sequence of products and services the user is interested in, determining all the emotion categories and corresponding bidirectional intensities involved in each product or service the user is interested in. This means matching the previously generated emotion categories and the intensity sequence of each emotion with the chronological order of the products and services the customer is interested in. Through this alignment, the specific emotion category accompanying the customer's attention to each product or service, as well as the intensity of these emotions in the positive and negative directions, can be clearly identified. For example, when a customer is interested in a mobile phone, the corresponding emotion is hesitation, and the intensity of hesitation is moderate; when interested in a phone case, the emotion is satisfaction, and the intensity of satisfaction is high. This determines the emotion category and bidirectional intensity (bidirectional means negative and positive, with negative values ​​represented by negative values ​​and positive values ​​represented by positive values) involved in each product or service of interest.

[0145] In this embodiment, a bidirectional interest assignment result is generated for each product service of interest based on all relevant emotion categories and their corresponding bidirectional intensities. The network distribution map of the user's interested products is then marked based on these bidirectional interest assignment results, resulting in a real-time product interest heatmap for store customers. Specifically, based on the customer's emotion category and intensity for each product service of interest, at least one emotion category and a numerical result reflecting the degree of customer interest are assigned to each product service; this is the bidirectional interest assignment result. For example, the bidirectional intensity values ​​of all emotion categories corresponding to each product service are retrieved from the list of corresponding emotion types to determine the corresponding bidirectional interest assignment. The bidirectional interest assignment values ​​of all emotion categories corresponding to each product service are then summed to obtain the bidirectional interest assignment result for a single product service of interest.

[0146] Then, these assigned values ​​are mapped onto the previously generated network distribution map of products of interest. Different colors, brightness levels, or other visual methods are used to mark the degree of interest in different products and services, ultimately forming a heatmap that intuitively displays the real-time distribution of customer product interests. For example, products with high customer interest are highlighted in red, while those with low interest are displayed in blue, making it easy for sales staff to quickly understand the customer's interest in different products.

[0147] The beneficial effects of the above technologies are as follows: The emotion category recognition unit analyzes changes and intensity of customer emotions, providing a quantitative basis for interest analysis. The two-way intensity determination unit aligns emotions with the product interest sequence in time, establishing a dynamic relationship between the two and fully considering the impact of the time dimension. The two-way value assignment and labeling unit labels the product network distribution map accordingly, generating a real-time product interest heatmap. This heatmap intuitively and accurately reflects the customer's interest in the product and their emotional attitude, helping sales personnel to understand customer psychology, adjust sales strategies accordingly, optimize sales scripts, improve the quality and success rate of sales interactions, and enhance the functionality of the store's real-time interactive auxiliary system.

[0148] Example 9: Based on Example 1, an advanced shopping guide assistance module is provided, see reference. Figure 4 ,include:

[0149] The thermal level zone molecular module is used to classify the real-time product interest heatmap into thermal levels and obtain the local real-time product interest local thermal map under each thermal level.

[0150] The advanced sales guide assistance submodule is used to input the local real-time product interest heatmaps under all heat levels and the latest needs of store customers into the preset AI model to generate advanced sales guide scripts until the sales guide task is completed.

[0151] In this embodiment, the real-time product interest heatmap is differentiated by heat level to obtain a local real-time product interest heatmap under each heat level. This means that the heatmap, which displays the level of customer interest in products in real time, is divided into different levels according to the intensity of interest, based on certain rules. For example, the area with the highest intensity is heat level one, the slightly lower area is heat level two, and so on. Each level corresponds to a local area, and these areas form the local real-time product interest heatmap under each heat level. This allows for a clearer display of the stratification of customer interest in different products.

[0152] In this embodiment, local real-time product interest heatmaps at all heat levels and the latest needs-related text from store customers are input into a pre-trained AI model to generate advanced sales scripts until the sales task is completed. This involves inputting the previously categorized local heatmaps at different heat levels, along with text extracted from the customer's latest statements related to their needs, into a pre-trained AI model. Based on this input information, the AI ​​model generates sales language content more closely aligned with the customer's current interests and needs—the advanced sales scripts. Throughout the sales process, this process is continuously repeated based on customer feedback until the customer makes a purchase decision, abandons the purchase, or otherwise concludes the sales task.

[0153] The beneficial effects of the above technologies are as follows: The heat map module categorizes real-time product interest heatmaps into heat maps of varying intensity levels, resulting in local heat maps for each level. This allows for a clearer and more detailed presentation of customer interest in products. Different heat maps correspond to different areas of interest intensity, facilitating subsequent precise analysis. The advanced sales guidance submodule uses these local heat maps and the latest customer needs, along with text input to a preset AI model, to generate advanced sales guidance scripts. By combining interest intensity distribution and latest needs, the AI ​​model can generate more targeted scripts, meeting customer needs under different interest states and improving the accuracy and effectiveness of sales guidance. This approach better guides customers to purchase products, enhances sales interaction, increases sales success rates, and improves the functionality of the store's real-time interactive support system.

[0154] Example 10: This invention provides a real-time interactive assistance method for store sales based on an AI agent, comprising:

[0155] S1: Extract the text related to customer needs from the real-time voice of customers in the store, and interpret the text related to customer needs to obtain customer needs.

[0156] S2: Connect to the CRM system to read store inventory information and use a preset AI model to generate initial sales scripts based on customer needs and store inventory information;

[0157] S3: Acquire emotional capture information and strong attention action recognition information of store customers during the process of receiving the initial sales guide script, and generate a real-time product interest heatmap of store customers based on the emotional capture information and strong attention action recognition information.

[0158] S4: Based on the latest needs of store customers, generate advanced sales scripts involving text and real-time product interest heatmaps until the sales task is completed.

[0159] The beneficial effects of the above technologies are as follows: Extracting and interpreting customer needs from real-time voice recordings accurately grasps customer intentions, laying the foundation for sales, avoiding communication errors, and enhancing the customer shopping experience. Integrating with a CRM system and combining inventory with AI models generates initial sales scripts, ensuring recommendations are based on actual inventory while making the scripts professional and accurate, thus improving sales efficiency. S3 acquires customer emotion and action information to generate product interest heatmaps, providing in-depth insights into changes in customer interests and offering strong support for subsequent sales guidance. Advanced sales scripts are generated based on the latest demand text and interest heatmaps, aligning with dynamic customer needs and achieving personalized and precise sales guidance.

[0160] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of this invention and its equivalents, this invention also intends to include these modifications and variations.

Claims

1. An AI agent-based real-time interactive assistant system for store sales, characterized in that, The method comprises the following steps: The demand interpretation module is used to extract the demand-related text of the store customer from the real-time voice of the store customer, interpret the demand-related text of the customer, and obtain the customer demand; The initial guide auxiliary module is used to access the CRM system to read the store inventory information, and use the preset AI model to generate the initial guide speech based on the customer demand and the store inventory information; The product interest analysis module is used to obtain the emotion capture information and strong attention action recognition information of the store customer during the process of receiving the initial guide speech, and generate the real-time product interest heat map of the store customer based on the emotion capture information and the strong attention action recognition information; The advanced guide auxiliary module is used to generate the advanced guide speech based on the latest demand-related text of the store customer and the real-time product interest heat map until the guide task is completed; The product interest analysis module comprises: The information capture submodule is used to read the emotion capture information and strong attention action recognition information of the store customer during the process of receiving the initial guide speech in the store monitoring video; The interested product distribution map generation submodule is used to determine the interested product service sequence of the user based on the strong attention action recognition information, and generate the interested product network distribution map of the user by combining the correlation degree between the store product services; The product interest heat map generation submodule is used to bidirectionally assign interest to the interested product network distribution map of the user based on the emotion capture information, and obtain the real-time product interest heat map of the store customer.

2. The AI agent-based store sales real-time interaction assistance system according to claim 1, wherein, The demand interpretation module comprises: The sentence and word segmentation processing submodule is used to convert the real-time voice of the store customer into voice text, and perform sentence and word segmentation processing on the voice text to obtain all customer input sentences and the word set of each customer input sentence; The vertical correlation degree determination submodule is used to determine the correlation degree between each word in each word set and the store product service as the vertical correlation degree of each word in each word set; The horizontal correlation degree determination submodule is used to obtain the horizontal correlation degree of each word in each word set based on the correlation degree between each word in each word set and all the remaining words in each word set except the current word; The demand-related two-dimensional submodule is used to take the vertical correlation degree and the horizontal correlation degree of each word as the vertical coordinate value and the horizontal coordinate value of the corresponding word in the preset two-dimensional coordinate system respectively, generate the mark point of each word under the preset coordinate system, and generate the demand-related two-dimensional graph based on the mark points of all words under the preset coordinate system; The demand-related text extraction submodule is used to extract the demand-related text of the store customer from the voice text based on the demand-related two-dimensional graph; The demand-related text interpretation submodule is used to interpret the demand-related text of the customer to obtain the customer demand. 3.The AI agent-based store sales real-time interaction assistance system of claim 2, wherein, The horizontal correlation degree determination submodule comprises: The order attribution division unit is used to take the correlation degree between each word in each word set and each remaining word in the corresponding word set except the current word as the same-order correlation degree of the corresponding two words, and take the correlation degree between each word in each word set and each word in all remaining word sets except the current word set as the different-order correlation degree of the corresponding two words. The tree structure building unit is configured to filter all strong correlation words from all words based on longitudinal correlation degrees of the all words, and build a strong correlation word development tree structure based on occurrence time sequences and original correlation relationships of the all strong correlation words; The word meaning aggregation unit is configured to take all the strong correlation words as clustering centers, and perform clustering and division on all words in all word sets based on correlation degrees between two words in the all word sets, to obtain a semantic aggregation word set of each strong correlation word. The transverse correlation degree determination unit is configured to determine a transverse correlation degree of each word in each word set based on the strong correlation word development tree structure, the semantic aggregation word set of the all strong correlation words, and the correlation degrees between two words in the all word sets. 4.The AI agent-based store sales real-time interaction assistance system of claim 3, wherein, The transverse correlation degree determination unit comprises: The first reinforcement calculation sub-unit is configured to determine a reinforcement factor of a same-order correlation degree between two words in the all word sets based on correlation degrees between the two words and strong correlation words of a corresponding semantic aggregation word set to which the two words belong when the two words belong to a same word set and a same semantic aggregation word set, and determine a reinforced correlation degree between the two words based on the same-order correlation degree and the corresponding reinforcement factor. The second reinforcement calculation sub-unit is configured to determine a reinforcement factor of a same-order correlation degree between two words in the all word sets based on correlation degrees between the two words and strong correlation words of respective semantic aggregation word sets to which the two words belong, correlation degrees between the strong correlation words of the respective semantic aggregation word sets, and a cross-set factor of the strong correlation words of the respective semantic aggregation word sets in the strong correlation word development tree structure when the two words belong to a same word set and do not belong to a same semantic aggregation word set, and determine a reinforced correlation degree between the two words based on the same-order correlation degree and the corresponding reinforcement factor. The third reinforcement calculation sub-unit is configured to determine a reinforcement factor of a different-order correlation degree between two words in the all word sets based on correlation degrees between the two words and strong correlation words of respective semantic aggregation word sets to which the two words belong, correlation degrees between the strong correlation words of the respective semantic aggregation word sets, and a cross-layer factor of the strong correlation words of the respective semantic aggregation word sets in the strong correlation word development tree structure when the two words do not belong to a same word set and do not belong to a same semantic aggregation word set, and determine a reinforced correlation degree between the two words based on the different-order correlation degree and the corresponding reinforcement factor. The transverse correlation degree determination sub-unit is configured to take a mean value of the reinforced correlation degrees between each word in each word set and all words in the all word sets except a current word as the transverse correlation degree of each word in each word set. 5.The AI agent based real-time interaction assistance system for store sales according to claim 2, wherein, The requirement-related text extraction submodule comprises: The division span value determination unit is configured to determine a distance between a nearest marker point to the origin and a farthest marker point to the origin in the requirement-related two-dimensional graph as a span value of the requirement-related two-dimensional graph, and take a product of the span value of the requirement-related two-dimensional graph and a preset proportion and a sum of a distance between the nearest marker point to the origin and the origin as a division span value of the requirement-related two-dimensional graph. The demand involves a mark point screening unit, which is used to divide a circular area in a preset coordinate system with a preset two-dimensional coordinate system as an origin and a division span value as a radius, and take all mark points outside the circular area in the demand-involved two-dimensional graph as all demand-involved mark points; The demand-involved text extraction unit is used to extract demand-involved text of a store customer from a voice text based on all demand-involved mark points in the demand-involved two-dimensional graph. 6.The AI agent-based store sales real-time interaction assistance system of claim 5, wherein, The demand-involved text extraction unit comprises: A cluster position marking subunit is configured to determine distribution positions of all mark points corresponding to each word set in a preset two-dimensional coordinate system as a cluster position of the word set; A duplicate demand judgment subunit is configured to calculate a similarity between cluster positions of each two word sets, and combine all customer input sentences corresponding to all word sets with a similarity exceeding a similarity threshold as a to-be-processed duplicate extraction text, and take each customer input sentence not combined as a to-be-processed extraction text; A duplicate extraction and extraction subunit is configured to perform duplicate extraction on the to-be-processed duplicate extraction text, and perform extraction on the to-be-processed extraction text to obtain demand-involved text of a store customer. 7.The AI-agent-based real-time interaction assistance system for store sales according to claim 1, wherein, The product interest heat map generation sub-module comprises: An emotion category identification unit is configured to generate an emotion category sequence of a store customer in a process of receiving an initial guide speech based on emotion capture information, and perform bidirectional intensity evaluation on each emotion in the emotion category sequence to obtain an emotion category and intensity sequence; A bidirectional intensity determination unit is configured to perform time sequence alignment between the emotion category and intensity sequence and a product and service sequence of interest of a user to determine all involved emotion categories and corresponding bidirectional intensities of each product and service of interest of the user; A bidirectional assignment marking unit is configured to generate a bidirectional interest assignment result of each product and service of interest based on all involved emotion categories and corresponding bidirectional intensities of each product and service of interest, mark a product interest network distribution graph of the user based on the bidirectional interest assignment results of all product and service of interest, and obtain a real-time product interest heat map of the store customer. 8.The AI-agent-based real-time interaction assistance system for store sales according to claim 1, wherein, The advanced guide assistance module comprises: A heat level differentiation sub-module is configured to differentiate heat levels of the real-time product interest heat map to obtain a local real-time product interest local heat map under each heat level; An advanced guide assistance sub-module is configured to input all local real-time product interest local heat maps under all heat levels and the latest demand-involved text of the store customer into a preset AI model to generate an advanced guide speech until a guide task is completed. 9.A method for real-time interaction assistance of store sales based on an AI agent, characterized in that, The method comprises: S1: extracting demand-involved text of a store customer from real-time voice of the store customer, and interpreting the demand-involved text of the customer to obtain customer demand; S2: accessing a CRM system to read store inventory information, and generating an initial guide speech based on customer demand and store inventory information by using a preset AI model; S3: obtaining emotion capture information and strong attention action recognition information of the store customer in a process of receiving the initial guide speech, and generating a real-time product interest heat map of the store customer based on the emotion capture information and the strong attention action recognition information; S4: Based on the latest needs of the store customers involving text and real-time product interest heat map generation advanced guide words, until the end of the guide task; Wherein, step S3: obtaining the emotion capture information and strong attention action recognition information of the store customers in the process of receiving the initial guide words, generating the real-time product interest heat map of the store customers based on the emotion capture information and the strong attention action recognition information, comprising: An information capture sub-module, configured to read the emotion capture information and strong attention action recognition information of the store customers in the process of receiving the initial guide words in the store monitoring video; An interested product distribution map generation sub-module, configured to determine the interested product service sequence of the user based on the strong attention action recognition information, and generate the interested product network distribution map of the user in combination with the correlation degree between the store product services; A product interest heat map generation sub-module, configured to perform interest bidirectional assignment on the interested product network distribution map of the user based on the emotion capture information, and obtain the real-time product interest heat map of the store customers.

Images