A conversational intelligent shopping guide method and system for the cultural, commercial and tourism sectors

Abstract

This invention discloses a conversational intelligent shopping guide method and system for the cultural, commercial, and tourism sectors. The core of this method lies in transforming the traditional "keyword matching response" of private domain services into "deep semantic understanding and multi-turn dialogue interaction." First, this method establishes a multi-turn dialogue management mechanism, achieving context awareness through an attention mechanism and automatically filling missing slots. Simultaneously, it employs a pre-trained large language model for deep semantic understanding and multi-tag intent classification, mapping user queries to standard entities in the brand knowledge graph through named entity recognition and entity linking. Then, based on user profiles and knowledge graph subgraph retrieval, it achieves multi-target personalized recommendations and generates natural language explanations for each recommendation to enhance user trust. Finally, through a service execution module, it completes a one-stop closed-loop service including coupon distribution, reservation ordering, and store navigation, and saves dialogue summaries to a user profile database to form accumulable data assets. This invention effectively solves the problem of experience fragmentation after public domain traffic enters the private domain, automating over 80% of core operational processes and significantly improving private domain traffic conversion rates and user experience satisfaction.

Description

Technical Field

[0001] This invention relates to the fields of artificial intelligence, natural language processing, and intelligent recommendation technology, specifically to a conversational intelligent shopping guide method and system for the cultural, commercial, and tourism sectors, applicable to scenarios of private domain traffic acquisition and intelligent service conversion in physical businesses. Background Technology

[0002] With the advent of the era of large-scale models, cultural, commercial, and tourism enterprises are acquiring a large amount of AI-driven traffic in the public domain through new technologies such as GEO (Generative Engine Optimization). However, when users move from the public domain (AI dialogue, content community) to the brand's private domain (mini-program, official website, app), traditional enterprise service systems face severe technical challenges.

[0003] First, the problem of fragmented interactive experience is prominent. Traditional search boxes or menu navigation cannot handle natural language conversational traffic. Users are used to free expression in the public domain, but they need to adapt to rigid operation logic when entering the private domain, resulting in a large loss of traffic.

[0004] Secondly, the understanding of intent is superficial. Existing customer service robots are mostly based on keyword matching or simple intent classification, which cannot understand the user's deep semantic needs, especially in handling complex scenario-based queries in the cultural, commercial and tourism fields (such as "Where is a good place to go for the elderly and children on the weekend?").

[0005] Secondly, the recommendation strategy is too simplistic. Traditional recommendation systems rely on collaborative filtering or tag matching, lacking in-depth utilization of brand knowledge graphs, and thus failing to achieve personalized recommendations based on user intent, context, and preferences.

[0006] Finally, the service loop breaks down. Service steps such as shopping guides, recommendations, appointments, and discount distribution are scattered across different systems, making it impossible to achieve a one-stop, closed-loop service through conversational interaction.

[0007] Therefore, the industry urgently needs an intelligent shopping guide technology solution that can deeply understand user needs through natural language dialogue, provide personalized recommendations based on knowledge graphs, and automate the entire service process. Summary of the Invention

[0008] The purpose of this invention is to overcome the shortcomings of the prior art and provide a conversational intelligent shopping guide method and system for the cultural, commercial and tourism fields.

[0009] 1. Core Technology Architecture

[0010] This invention establishes a complete chain in its architecture: "dialogue interaction – semantic understanding – knowledge query – personalized recommendation – service execution," and includes the following specific steps:

[0011] S1: Multi-turn Dialogue Management. When a user enters the private dialogue interface, the system loads the user's historical profile and a subset of the brand's knowledge graph, initializing the dialogue state. In each round of dialogue, the system maintains the dialogue history and calculates the relevance weight between past rounds and the current input through an attention mechanism, achieving context-aware response generation. For queries lacking key information, the system automatically initiates clarifying questions, filling in necessary slots (such as time, number of people, budget, preferences, etc.) through multiple rounds of interaction. When the user clearly expresses their intention to end the session or completes the service loop, the system automatically saves the dialogue summary to the user profile database.

[0012] S2: Deep Semantic Understanding and Intent Recognition. A pre-trained large language model is used to encode user input and extract semantic vector representations. Multi-label intent classification is performed based on these semantic vectors to identify the user's primary intent (e.g., inquiry, comparison, reservation, complaint) and secondary intent. Named Entity Recognition (NER) technology is used to extract entities such as brand name, product name, location, and time from user queries and map them to standard entities in a knowledge graph through entity linking technology. Simultaneously, the sentiment and urgency of user input are analyzed to adjust response strategies and service priorities.

[0013] S3: Knowledge Graph Query and Personalized Recommendation. Based on the identified user intent and profile tags, relevant subgraphs are retrieved from the brand knowledge graph to obtain a set of candidate recommendations. Considering multiple optimization objectives such as conversion rate, average order value, user satisfaction, and inventory status, the candidate recommendations are ranked according to multiple objectives. Natural language explanations are generated for each recommendation, explaining the reasons for the recommendation (e.g., "Based on your preference for designer collaborations, we recommend this KAWS collaboration UT"), improving user trust and click-through rates.

[0014] S4: Service Execution and Closure. Based on user confirmation, execute service actions such as coupon distribution, order placement, information push, and store navigation. The system automatically saves dialogue summaries to the user profile database for subsequent personalized optimization. This achieves full-process automation from "dialogue interaction" to "service completion."

[0015] 2. Beneficial effects

[0016] Enhanced user experience: Through human-like dialogue interaction, user satisfaction is increased by more than 40% compared to traditional search, and the dialogue completion rate reaches over 85%. Users can continue their natural language dialogue habits in their private domain, receiving a smooth shopping guide service experience.

[0017] Improve conversion rates: Through precise intent recognition and personalized recommendations, private domain traffic conversion rates are 2-3 times higher than traditional search. Natural language interpretation mechanisms enhance user trust in recommendation results.

[0018] Reduce operating costs: Automate over 80% of core operational processes, significantly reducing the need for human customer service. The automatic slot filling mechanism for multi-turn conversations minimizes the need for manual intervention.

[0019] Accumulate data assets: Each conversation enriches the user profile, forming an operational brand private domain data asset that supports long-term refined operations. User preference tags, spending power, scenario preferences, and other dimensions are continuously accumulated. Attached Figure Description

[0020] Figure 1 This is a general flowchart of the method of the present invention;

[0021] Figure 2 A schematic diagram illustrating multi-turn dialogue management and context awareness;

[0022] Figure 3 Flowchart for semantic understanding and intent recognition;

[0023] Figure 4 A diagram illustrating the logical architecture of a personalized recommendation engine;

[0024] Figure 5 This is a schematic diagram of a cross-platform deployment architecture and implementation examples. Detailed Implementation

[0025] The technical solution of the present invention will be described in detail below with reference to specific embodiments.

[0026] Example 1: Bicester "Henry" Smart Assistant

[0027] This embodiment deploys the "Henry" smart assistant at Bicester Village to verify the offline customer conversion effect of conversational shopping guides.

[0028] Dialogue Example 1 – Product Recommendation Scenario:

[0029] Users enter the "Henry" chat interface through a WeChat mini-program and type "I want to buy a pair of sneakers, budget around 1000 yuan." The system, through its semantic understanding module, identifies the user's intent as "product recommendation," extracts the entities "sneakers" and "budget 1000 yuan," and initiates clarification: "Do you prefer sports brands or casual brands?" The user replies "sports brands, Nike or Adidas." The system completes the slots and queries the knowledge graph to obtain brand store information and product SKUs matching the budget. The personalized recommendation engine generates a recommendation list based on the user profile (young male, preference for sportswear), along with store navigation and coupons. When the user clicks on "Nike Air Max" in the recommendation card, the system issues a limited-time coupon and guides the user to the store, completing the closed-loop conversion from chat to in-store visit.

[0030] Dialogue Example 2 – Scenario Consultation Scenario:

[0031] A user asked, "What recommendations do you have for bringing elderly people and children this weekend?" The system identified the intent as a "scenario-based consultation," extracted the entities "elderly," "children," and "weekend," and inferred the implicit need to be family-friendly activities. It then queried a knowledge graph to obtain information on family-friendly services such as facilities for children, accessible pathways, and rest areas, and combined this with restaurant recommendations to generate a personalized itinerary. The system proactively asked if the user needed to book VIP services or childcare, transforming a simple inquiry into a deeper service interaction.

[0032] Example 2: Uniqlo's official website intelligent search integration

[0033] This embodiment deploys a conversational intelligent search on the Uniqlo website to leverage traffic from the GEO public domain.

[0034] When a user enters the Uniqlo website through an AI-recommended link, a greeting message automatically pops up in the chat window: "Hello! We see you are interested in the KAWS collaboration series. Would you like to learn more?" The system identifies that the user's source is the KAWS collaboration GEO and automatically loads a subset of the IP's knowledge graph and user profile.

[0035] When a user asks "What other collaboration items are still available?", the system queries the knowledge graph and returns a list of all currently available IP collaboration items, including those from KAWS, Takashi Murakami, and Jil Sander, and generates a recommendation reason for each item.

[0036] When a user further inquires about the availability of the KAWS collaboration item, the system checks the inventory in real time, provides the specific size and stock status, and recommends similar styles as alternatives. Once the user expresses their intention to purchase, the system directly redirects to the product details page and automatically fills in a coupon code, achieving a seamless purchasing experience.

Claims

1. A conversational intelligent shopping guide method for the cultural, commercial, and tourism sectors, characterized in that: include: Multi-turn dialogue management steps: When a user enters the private domain dialogue interface, the system loads the user's historical profile and a subset of the brand knowledge graph to initialize the dialogue state; In each round of dialogue, the relevance weight between the previous rounds and the current input is calculated through an attention mechanism to achieve context-aware response generation; for queries that lack key information, clarification questions are automatically initiated, and necessary slots are filled in through multiple rounds of interaction; The deep semantic understanding and intent recognition steps are as follows: user input is encoded using a pre-trained large language model to extract semantic vector representations; multi-label intent classification is performed based on semantic vectors to identify the user's explicit and implicit intents; named entity recognition technology is used to extract entities such as brand name, product name, location, and time from the user query, and these entities are mapped to standard entities in the knowledge graph through entity linking technology. Knowledge graph query and personalized recommendation steps: Based on the identified user intent and profile tags, retrieve relevant subgraphs in the brand knowledge graph to obtain a set of candidate recommendation items; comprehensively consider multiple optimization objectives such as conversion rate, average order value, user satisfaction, and inventory status to sort the candidate recommendation items according to multiple objectives; generate natural language explanations for each recommendation result to explain the reasons for the recommendation; Service execution and closed-loop steps: Based on the user's confirmation, perform service actions such as coupon distribution, order placement, information push, and store navigation; when the user clearly expresses the intention to end the service or completes the service loop, the system automatically saves the dialogue summary to the user profile database for subsequent personalized optimization.

2. The method according to claim 1, characterized in that: In the multi-round dialogue management steps, the necessary slots include time, number of people, budget, preferences, and scenario type. The system automatically generates clarification questions based on missing slots and infers the user's implicit needs through the context of the dialogue history.

3. The method according to claim 1, characterized in that: In the deep semantic understanding and intent recognition steps, the system also performs sentiment and urgency analysis to analyze the sentiment and urgency of the user input, and adjusts the response strategy and service priority accordingly, giving priority to high-urgency requests.

4. The method according to claim 1, characterized in that: In the knowledge graph query and personalized recommendation steps, the multi-objective ranking adopts a weighted scoring model, and the weight of each optimization objective is dynamically adjusted according to the current dialogue scenario. The natural language interpretation is automatically generated based on entity relationship paths in the knowledge graph, enhancing users' trust in the recommendation results.

5. The method according to claim 1, characterized in that: In the service execution and closed-loop steps, the system automatically updates the user profile's preference tags, spending power, and scenario preference dimensions based on the dialogue completion status, forming an accumulable user data asset that supports long-term refined operation.

6. A conversational intelligent shopping guide system for the cultural, commercial, and tourism sectors, characterized in that: include: Multi-turn dialogue management module: used for dialogue state tracking, context management, and dialogue flow control; Semantic understanding and intent recognition module: used to perform deep semantic analysis on user input and identify user intent and entity information; Knowledge graph query module: used to query the brand knowledge graph based on the recognition results to obtain accurate business data; Personalized recommendation engine module: used to combine user profiles and knowledge graph data to generate personalized recommendation results; Service execution module: Used for executing service actions such as coupon distribution, appointment order placement, information push, and store navigation; Cross-platform access module: Used to support rapid access and unified services for WeChat Mini Programs, Web, and App.

Images