Interactive automobile boutique recommendation method and system based on artificial intelligence

Abstract

The invention relates to the technical field of commodity recommendation, in particular to an interactive automobile boutique recommendation method and system based on artificial intelligence, and the system comprises a user interaction module which is used for capturing user demands through a multi-mode interaction mode and supporting real-time feedback and dynamic guidance; the data acquisition and preprocessing module is used for acquiring data input and covering acquisition, cleaning and standardization of user data and vehicle data; the user portrait construction module is used for generating a multi-dimensional user portrait based on the user data and accurately describing dominant and implicit demands; the vehicle feature extraction module is used for integrating vehicle tags contained in the vehicles to obtain vehicle tag features and construct a vehicle feature library; and the AI recommendation algorithm module is used for accurately matching the user portrait with the vehicle characteristics to generate a recommended vehicle. According to the method, the display and implicit requirements can be fused, the user portraits can be dynamically updated, the recommendation weight can be dynamically optimized according to the user satisfaction, and the recommendation effect is improved.

Description

Technical Field

[0001] This invention relates to the field of product recommendation technology, specifically to an interactive automotive product recommendation method and system based on artificial intelligence. Background Technology

[0002] Vehicle recommendation systems are tools to help users select vehicles, but existing systems have significant drawbacks: interaction relies on single text or forms, lacks multimodal support such as voice, struggles to handle ambiguous needs, and requires restarting after interruption; user profiles only contain explicit needs, ignoring implicit preferences such as browsing behavior, and updates are lagging; recommendations rely on simple tag comparisons, resulting in low accuracy and difficulty in optimizing based on feedback; these factors lead to a large deviation between recommendations and actual needs, thus requiring AI-based systems to improve effectiveness through multimodal interaction, dynamic profiling, and intelligent matching. Summary of the Invention

[0003] The purpose of this invention is to address the problems existing in the background technology by proposing an interactive car recommendation method and system based on artificial intelligence.

[0004] The technical solution of this invention: an interactive car recommendation system based on artificial intelligence, comprising:

[0005] The user interaction module is used to capture user needs through multimodal interaction and supports real-time feedback and dynamic guidance.

[0006] The user profile building module generates multi-dimensional user profiles based on user data, accurately depicting explicit and implicit needs;

[0007] The vehicle feature extraction module is used to integrate the vehicle tags contained in the vehicle to obtain vehicle tag features and build a vehicle feature library.

[0008] The AI ​​recommendation algorithm module is used to accurately match user profiles with vehicle features to generate recommended vehicles.

[0009] Preferably, the user interaction module includes a multimodal demand input unit, a multi-turn dynamic interaction unit, and an interaction state management unit;

[0010] The multimodal demand input unit is used to acquire the user's text and voice input, convert it into text demand information, combine natural language processing technology to parse key information, extract key information through the BERT pre-trained model and classify it into predefined vehicle labels, and integrate them to obtain an explicit demand label list.

[0011] It provides structured forms and visual controls for users to directly select their specific needs; for example, visual controls can be sliders and checkboxes; specific needs can be budget range, vehicle type, power type, brand preference;

[0012] Multi-round dynamic interaction units are used to perceive the ambiguity of textual requirements and make progressive questions based on the degree of ambiguity to refine the requirements;

[0013] The interaction state management unit is used to save the user's historical interaction records, including completed interactions and interrupted interactions; when an interrupted interaction is re-entered, the interaction continues from the last ending point.

[0014] The preferred formula for calculating ambiguity is: In the formula, M is the ambiguity; YLi is the predefined vehicle label obtained from classification; αi is the information richness quantification coefficient of the vehicle label; i is the vehicle label number, i is a positive integer, i∈[1,n], n is the total number of vehicle labels obtained from classification; TL is the text length.

[0015] Preferably, the user profile building module includes an explicit needs profile unit, an implicit needs mining unit, and a profile building dynamic update module;

[0016] The explicit demand profile unit is used to obtain users' car usage demand information; it is obtained through text and voice input.

[0017] The implicit demand mining unit is used to obtain user behavior information, including browsing history and search history, extract browsing page keywords and search keywords, integrate them into behavior information, and perform inference to generate preference tags.

[0018] Preferably, the reasoning includes calculating the frequency of each keyword in the behavioral information, comparing it with a preset frequency threshold, and marking keywords with a frequency greater than the frequency threshold as behavioral keywords;

[0019] K-means was used to cluster behavioral keywords into vehicle tags, and the clustering results were integrated to obtain a list of implicit demand tags.

[0020] Preferably, the user profile building module is also connected to a profile dynamic update unit, which is used to determine the user profile based on the explicit demand tag list and the implicit demand tag list, and update the profile tags in real time after each round of interaction with the user.

[0021] Preferred methods for accurately matching user profiles with vehicle characteristics include:

[0022] One-hot encoding is performed on vehicle tags. The user profile and vehicle tag features in the vehicle feature library are converted into vector form and the vector dimension is unified to obtain user profile vector and vehicle feature vector. Based on the user profile vector and all vehicle feature vectors in the vehicle feature library, cosine similarity is calculated to match the optimal vehicle.

[0023] Preferably, the formula for calculating cosine similarity is:

[0024] In the formula, YHj and CTj are the dimension values ​​of the user profile vector and the vehicle feature vector, respectively; βj is the vehicle label weight corresponding to the dimension, which is obtained by fitting based on historical big data; j is the dimension number, j is a positive integer, j∈[1,m], and m is the total number of vector dimensions;

[0025] The vehicles are sorted in descending order of cosine similarity to obtain a vehicle ranking list. The top three vehicles are then selected as the best vehicles for recommendation.

[0026] Preferably, for each recommended vehicle, a satisfaction signal is obtained, including both satisfied and dissatisfied responses. The number of satisfied and dissatisfied responses for multiple recent recommended vehicles are counted and labeled Sk and Dk respectively, where k is the vehicle number and k is a positive integer; based on the formula... The recommendation score Mk for each vehicle is calculated.

[0027] The recommendation score Mk of a vehicle is used as the correction coefficient Xj for each vehicle tag j it contains, and the correction weight is calculated based on the correction coefficient to update the vehicle tag weight. This achieves the goal of using user satisfaction as a reward signal to approach the optimal recommendation in multiple rounds of interaction.

[0028] Through expressions The corrected weight βj was calculated. 修正 In the formula, X1 and X2 are the preset first correction threshold and second correction threshold, respectively; where X1 < 0 <X2。

[0029] This invention also discloses an interactive car recommendation method based on artificial intelligence, which applies the aforementioned interactive car recommendation system based on artificial intelligence, and specifically includes the following steps:

[0030] Capture user needs through multimodal interaction methods and support real-time feedback and dynamic guidance;

[0031] User data generates multi-dimensional user profiles, accurately depicting explicit and implicit needs;

[0032] The vehicle tags contained in the vehicle are integrated to obtain vehicle tag features, and a vehicle feature library is constructed.

[0033] The system accurately matches user profiles with vehicle characteristics to generate recommended vehicles.

[0034] Compared with the prior art, the above-mentioned technical solution of the present invention has the following beneficial technical effects:

[0035] This invention addresses ambiguous needs and resumes interrupted interactions through multimodal input and dynamic questioning, resulting in a more intelligent interaction method. By integrating explicit and implicit needs and dynamically updating user profiles, it provides a more comprehensive user profile. Furthermore, it achieves precise matching through vector matching and cosine similarity calculation, and can dynamically optimize recommendation weights based on user satisfaction to improve recommendation effectiveness. Attached Figure Description

[0036] Figure 1 This is a layer structure diagram of Embodiment 1 of the present invention. Detailed Implementation

[0037] Example 1, as Figure 1 As shown, the interactive car recommendation system based on artificial intelligence proposed in this invention includes:

[0038] The user interaction module is used to capture user needs through multimodal interaction and supports real-time feedback and dynamic guidance.

[0039] The user interaction module includes a multimodal demand input unit, a multi-turn dynamic interaction unit, and an interaction state management unit.

[0040] The multimodal demand input unit is used to acquire the user's text and voice input, convert it into text demand information, combine natural language processing technology to parse key information, extract key information through the BERT pre-trained model and classify it into predefined vehicle labels, and integrate them to obtain an explicit demand label list.

[0041] It provides structured forms and visual controls for users to directly select their specific needs; for example, visual controls can be sliders and checkboxes; specific needs can be budget range, vehicle type, power type, brand preference;

[0042] Multi-round dynamic interaction units are used to perceive the ambiguity of textual requirements and make progressive questions based on the degree of ambiguity to refine the requirements;

[0043] The formula for calculating ambiguity is: In the formula, M is the ambiguity; YLi is the predefined vehicle label obtained from the classification; αi is the information richness quantification coefficient of the vehicle label; i is the vehicle label number, i is a positive integer, i∈[1,n], n is the total number of vehicle labels obtained from the classification; TL is the text length; where the unit of text length is bits;

[0044] The interaction state management unit is used to save the user's historical interaction records, including completed interactions and interrupted interactions; when an interrupted interaction is re-entered, the interaction continues from the last ending point;

[0045] The user profile building module generates multi-dimensional user profiles based on user data, accurately depicting explicit and implicit needs;

[0046] The user profile building module includes an explicit needs profiling unit and an implicit needs mining unit;

[0047] The explicit demand profile unit is used to obtain users' car usage demand information; it is obtained through text and voice input.

[0048] The implicit demand mining unit is used to obtain user behavior information, including browsing history and search history, extract browsing page keywords and search keywords, integrate them into behavior information, and perform inference to generate preference tags;

[0049] Reasoning involves calculating the frequency of each keyword in the behavioral information, comparing it with a preset frequency threshold, and marking keywords with a frequency greater than the frequency threshold as behavioral keywords.

[0050] K-means was used to cluster behavioral keywords into vehicle tags, and the clustering results were integrated to obtain a list of implicit demand tags;

[0051] The dynamic profile update unit is used to determine the user profile based on the list of explicit demand tags and the list of implicit demand tags, and to update the profile tags in real time after each round of interaction with the user.

[0052] The vehicle feature extraction module is used to integrate the vehicle tags contained in the vehicle to obtain vehicle tag features and build a vehicle feature library.

[0053] The AI ​​recommendation algorithm module is used to accurately match user profiles with vehicle features to generate recommended vehicles.

[0054] Methods for accurately matching user profiles with vehicle characteristics include:

[0055] One-hot encoding is performed on vehicle tags. The user profile and vehicle tag features in the vehicle feature library are converted into vector form and the vector dimension is unified to obtain user profile vector and vehicle feature vector. Based on the user profile vector and all vehicle feature vectors in the vehicle feature library, cosine similarity is calculated to match the optimal vehicle.

[0056] The formula for calculating cosine similarity is:

[0057] In the formula, YHj and CTj are the dimension values ​​of the user profile vector and the vehicle feature vector, respectively; βj is the vehicle label weight corresponding to the dimension, which is obtained by fitting based on historical big data; j is the dimension number, j is a positive integer, j∈[1,m], and m is the total number of vector dimensions;

[0058] The vehicles are sorted in descending order of cosine similarity to obtain a vehicle ranking list. The top three vehicles are selected as the best vehicles for recommendation.

[0059] For each recommended vehicle, a satisfaction signal is obtained, including both satisfied and dissatisfied responses. The number of satisfied and dissatisfied responses for multiple recent vehicle recommendations is counted and labeled Sk and Dk, respectively, where k is the vehicle ID and k is a positive integer. Based on the formula... The recommendation score Mk for each vehicle is calculated.

[0060] The recommendation score Mk of a vehicle is used as the correction coefficient Xj for each vehicle tag j it contains, and the correction weight is calculated based on the correction coefficient to update the vehicle tag weight. This achieves the goal of using user satisfaction as a reward signal to approach the optimal recommendation in multiple rounds of interaction.

[0061] Through expressions The corrected weight βj was calculated. 修正 In the formula, X1 and X2 are the preset first correction threshold and second correction threshold, respectively; where X1 < 0 <X2;

[0062] By using multimodal input and dynamic questioning, it can handle ambiguous needs and resume interrupted interactions, making the interaction method more intelligent; by integrating explicit and implicit needs and dynamically updating user profiles, it makes user profiles more comprehensive; by using vector matching and cosine similarity calculation to achieve precise matching, it can also dynamically optimize recommendation weights based on user satisfaction, thereby improving recommendation performance.

[0063] Example 2: The AI-based interactive car recommendation method proposed in this invention is applied to the AI-based interactive car recommendation system proposed in Example 1, and specifically includes the following steps:

[0064] Capture user needs through multimodal interaction methods and support real-time feedback and dynamic guidance;

[0065] User data generates multi-dimensional user profiles, accurately depicting explicit and implicit needs;

[0066] The vehicle tags contained in the vehicle are integrated to obtain vehicle tag features, and a vehicle feature library is constructed.

[0067] The system accurately matches user profiles with vehicle characteristics to generate recommended vehicles.

[0068] The embodiments of the present invention have been described in detail above with reference to the accompanying drawings. However, the present invention is not limited thereto. Various changes can be made within the scope of knowledge possessed by those skilled in the art without departing from the spirit of the present invention.

Claims

1. An interactive car recommendation system based on artificial intelligence, characterized in that, include: The user interaction module is used to capture user needs through multimodal interaction and supports real-time feedback and dynamic guidance. The user profile building module generates multi-dimensional user profiles based on user data, accurately depicting explicit and implicit needs; The vehicle feature extraction module is used to integrate the vehicle tags contained in the vehicle to obtain vehicle tag features and build a vehicle feature library. The AI ​​recommendation algorithm module is used to accurately match user profiles with vehicle features to generate recommended vehicles.

2. The AI-based interactive car recommendation system according to claim 1, characterized in that, The user interaction module includes a multimodal demand input unit, a multi-turn dynamic interaction unit, and an interaction state management unit. The multimodal demand input unit is used to acquire the user's text and voice input, convert it into text demand information, combine natural language processing technology to parse key information, extract key information through the BERT pre-trained model and classify it into predefined vehicle labels, and integrate them to obtain an explicit demand label list. It provides structured forms and visual controls for users to directly select their specific needs; for example, visual controls can be sliders and checkboxes; specific needs can be budget range, vehicle type, power type, brand preference; Multi-round dynamic interaction units are used to perceive the ambiguity of textual requirements and make progressive questions based on the degree of ambiguity to refine the requirements; The interaction state management unit is used to save the user's historical interaction records, including completed interactions and interrupted interactions; when an interrupted interaction is re-entered, the interaction continues from the last ending point.

3. The AI-based interactive car recommendation system according to claim 2, characterized in that, The formula for calculating ambiguity is: In the formula, M is the ambiguity; YLi is the predefined vehicle label obtained from classification; αi is the information richness quantification coefficient of the vehicle label; i is the vehicle label number, i is a positive integer, i∈[1,n], n is the total number of vehicle labels obtained from classification; TL is the text length.

4. The AI-based interactive car recommendation system according to claim 2, characterized in that, The user profile building module includes an explicit needs profile unit, an implicit needs mining unit, and a profile building dynamic update module; The explicit demand profile unit is used to obtain users' car usage demand information; it is obtained through text and voice input. The implicit demand mining unit is used to obtain user behavior information, including browsing history and search history, extract browsing page keywords and search keywords, integrate them into behavior information, and perform inference to generate preference tags.

5. The AI-based interactive car recommendation system according to claim 4, characterized in that, Reasoning involves calculating the frequency of each keyword in the behavioral information, comparing it with a preset frequency threshold, and marking keywords with a frequency greater than the frequency threshold as behavioral keywords. K-means was used to cluster behavioral keywords into vehicle tags, and the clustering results were integrated to obtain a list of implicit demand tags.

6. The AI-based interactive car recommendation system according to claim 5, characterized in that, The user profile building module is also connected to a profile dynamic update unit, which is used to determine the user profile based on the list of explicit demand tags and the list of implicit demand tags, and update the profile tags in real time after each round of interaction with the user.

7. The AI-based interactive car recommendation system according to claim 6, characterized in that, Methods for accurately matching user profiles with vehicle characteristics include: One-hot encoding is performed on vehicle tags. The user profile and vehicle tag features in the vehicle feature library are converted into vector form and the vector dimension is unified to obtain user profile vector and vehicle feature vector. Based on the user profile vector and all vehicle feature vectors in the vehicle feature library, cosine similarity is calculated to match the optimal vehicle.

8. The AI-based interactive car recommendation system according to claim 7, characterized in that, The formula for calculating cosine similarity is: In the formula, YHj and CTj are the dimension values ​​of the user profile vector and the vehicle feature vector, respectively; βj is the vehicle label weight corresponding to the dimension, which is obtained by fitting based on historical big data; j is the dimension number, j is a positive integer, j∈[1,m], and m is the total number of vector dimensions; The vehicles are sorted in descending order of cosine similarity to obtain a vehicle ranking list. The top three vehicles are then selected as the best vehicles for recommendation.

9. The AI-based interactive car recommendation system according to claim 8, characterized in that, For each recommended vehicle, a satisfaction signal is obtained, including both satisfied and dissatisfied responses. The number of satisfied and dissatisfied responses for multiple recent vehicle recommendations is counted and labeled Sk and Dk, respectively, where k is the vehicle ID and k is a positive integer. Based on the formula... The recommendation score Mk for each vehicle is calculated. The recommendation score Mk of a vehicle is used as the correction coefficient Xj for each vehicle tag j it contains, and the correction weight is calculated based on the correction coefficient to update the weight of the vehicle tag. This achieves the goal of using user satisfaction as a reward signal to approach the optimal recommendation in multiple rounds of interaction. Through expressions The corrected weight βj was calculated. 修正 In the formula, X1 and X2 are the preset first correction threshold and second correction threshold, respectively; where X1 < 0 <X2。 10. An AI-based interactive car recommendation method, applied to the AI-based interactive car recommendation system described in any one of claims 1 to 9, characterized in that, Specifically, the following steps are included: Capture user needs through multimodal interaction methods and support real-time feedback and dynamic guidance; User data generates multi-dimensional user profiles, accurately depicting explicit and implicit needs; The vehicle tags contained in the vehicle are integrated to obtain vehicle tag features, and a vehicle feature library is constructed. The system accurately matches user profiles with vehicle characteristics to generate recommended vehicles.

Images