Vehicle recommendation system

Abstract

The present invention relates to a system (1) for recommending vehicles that best suit the needs of users by analysing past behaviour of users and specified vehicle features.

Description

[0001] DESCRIPTION

[0002] VEHICLE RECOMMENDATION SYSTEM

[0003] Technical Field

[0004] The present invention relates to a system for recommending vehicles that best suit the needs of users by analysing past behaviour of users and specified vehicle features.

[0005] Background of the Invention

[0006] Among the known vehicle recommendation techniques, content-based filtering, collaborative filtering and hybrid systems stand out. In content-based filtering, the features of items that users have liked or been interested in the past are analysed and new items with similar features are recommended. In this method, the content of items is represented by using contextual embedding techniques in the form of CountVectorizer, TF-IDF and Word2Vec. However, among the limitations of this approach is that users are limited to their previous fields of interest and are insufficient in exploring new fields of interest. Collaborative filtering, on the other hand, makes recommendations by associating similar users or products through previous interactions of users. Techniques such as Matrix Factorization, Singular Value Decomposition (SVD), Probabilistic Matrix Factorization (PMF) and Nonnegative Matrix Factorization (NMF) are used in order to discover hidden patterns. While this method provides recommendations by taking into account similarities between users, it may face problems such as lack of data or user-based biases. Hybrid systems aim to overcome these limitations by combining both approaches; however, additional challenges may arise in terms of system complexity and data management. For this reason, it is understood that there is a need for a system for recommending vehicles that best suit the needs of users by analysing past behaviour of users and specified vehicle features.

[0007] The Korean patent document no. KR20240097268, an application included in the state of the art, discloses a system for data implementation of customers and vehicle purchase recommendations based on big data and the operation method thereof. The said invention comprises a server; a processor; a communication module; a memory containing instructions, which, when executed by a processor, cause the processor to: receive data including consumption patterns and how the user purchases a vehicle; to enter into the vehicle selection model the user's preferred vehicle, age, previously purchased vehicle information and previously purchased vehicle information of other users in the user's age range, stored in the memory, in order to select the first plurality of vehicles among the vehicle information stored in the memory and to select information; to cluster the first plurality of vehicle information according to the intended use of the vehicle and the method of vehicle purchase, and to select the second plurality of vehicle information according to the user's intended use of the vehicle and the method of vehicle purchase; and can be adjusted to run the step of selecting at least one vehicle information that matches a consumption pattern among a second plurality of vehicle information including the vehicle price, the credit rating of the user and the monthly expenditure amount of the user.

[0008] Summary of the Invention

[0009] An object of the present invention is to realize a system developed with the aim of recommending vehicles that best suit the needs of users by analysing past behaviour of users and specified vehicle features. Another object of the present invention is to realize a system developed with the aim of overcoming the shortcomings of existing methods and meeting specific needs in the automotive sector.

[0010] A further object of the present invention is to realize a system developed with the aim of improving the user experience and increasing sales in the automotive sector by providing personalized recommendations suited to the individual needs and preferences of users.

[0011] Detailed Description of the Invention

[0012] “Vehicle Recommendation System” realized to fulfil the objectives of the present invention is shown in the figure attached, in which:

[0013] Figure l is a schematic view of the inventive system.

[0014] The components illustrated in the figure are individually numbered, where the numbers refer to the following:

[0015] 1. System

[0016] 2. Application

[0017] 3. Server

[0018] The inventive system (1) developed with the aim of recommending vehicles that best suit the needs of users by analysing past behaviour of users and specified vehicle features comprises; at least one application (2) which is configured to enable users to receive vehicle recommendations by entering specific criteria through an interface; and at least one server (3) which is configured to enable contextual embedding representations to be generated by using vehicle features, listing information and expertise information; similarity between features to be measured by using cosine similarity and functional features to be weighted according to the determined importance levels; data matrices to be divided into lower dimensional matrices and hidden relationships to be discovered by using the activity data of the users via matrix factorization techniques; and recommendations to be transmitted to the users through the application (2) according to the preferences and features of the users via tree-based ranking algorithms.

[0019] The application (2) included in the inventive system (1) is configured to communicate and exchange data with the server (3) by using any communication protocol. The application (2) is configured to enable the display of the attributes on which the recommendations provided to the user are based, in a comprehensible way, through an interface.

[0020] The server (3) included in the inventive system (1) is configured to communicate and exchange data with the application (2) by using any communication protocol. The server (3) is configured to work in two stages by combining content-based and collaborative filtering techniques. The server (3) is configured to enable contextual embedding representations to be generated by using vehicle features, listing information and expertise information via the content-based BERT model; similarity between features to be measured by using cosine similarity and functional features to be weighted according to the determined importance levels. The server (3) is configured to enable data matrices to be divided into lower dimensional matrices and hidden relationships to be discovered by using the activity data of users via the Neural Matrix Factorization technique. The server (3) is configured to enable recommendations to be ranked according to the preferences and features of users by using tree-based ranking algorithms. The server (3) is configured enable the reasons for the recommendations to be explained to the users by determining the contribution of each attribute to the predictions of the model via the SHAP (Shapley Additive Explanations) technique. The server (3) is configured to enable the recommendation list generated by the collaborative filtering technique to be customized for the user via a tree-based ranking algorithm; the decision-making processes of the algorithm to be made transparent via the SHAP technique; and the features on the basis of which the recommendations provided to the users are selected to be provided to the users with an easily comprehensible visualization through the application (2) interface. The server (3) is configured to enable diverse and rich content to be provided to users by integrating the collaborative filtering technique into contentbased recommendations.

[0021] Industrial Application of the Invention

[0022] By means of the inventive system (1), it is enabled to recommend vehicles that best suit the needs of users by analysing past behaviour of users and specified vehicle features.

[0023] Within these basic concepts; it is possible to develop various embodiments of the inventive “Vehicle Recommendation System (1)”; the invention cannot be limited to examples disclosed herein and it is essentially according to claims.

Claims

CLAIMS1. A system (1) for recommending vehicles that best suit the needs of users by analysing past behaviour of users and specified vehicle features comprising; at least one application (2) which is configured to enable users to receive vehicle recommendations by entering specific criteria through an interface; and characterized by at least one server (3) which is configured to enable contextual embedding representations to be generated by using vehicle features, listing information and expertise information; similarity between features to be measured by using cosine similarity and functional features to be weighted according to the determined importance levels; data matrices to be divided into lower dimensional matrices and hidden relationships to be discovered by using the activity data of the users via matrix factorization techniques; and recommendations to be transmitted to the users through the application (2) according to the preferences and features of the users via tree-based ranking algorithms.

2. A system (1) according to Claim 1; characterized by the application (2) which is configured to communicate and exchange data with the server (3) by using any communication protocol.

3. A system (1) according to Claim 1 or 2; characterized by the application (2) which is configured to enable the display of the attributes on which the recommendations provided to the user are based, in a comprehensible way, through an interface.

4. A system (1) according to any one of the preceding claims; characterized by the server (3) which is configured to communicate and exchange data with the application (2) by using any communication protocol.

5. A system (1) according to any one of the preceding claims; characterized by the server (3) which is configured to work in two stages by combining contentbased and collaborative filtering techniques.

6. A system (1) according to any one of the preceding claims; characterized by the server (3) which is configured to enable contextual embedding representations to be generated by using vehicle features, listing information and expertise information via the content-based BERT model; similarity between features to be measured by using cosine similarity and functional features to be weighted according to the determined importance levels.

7. A system (1) according to any one of the preceding claims; characterized by the server (3) which is configured to enable data matrices to be divided into lower dimensional matrices and hidden relationships to be discovered by using the activity data of users via the Neural Matrix Factorization technique.

8. A system (1) according to any one of the preceding claims; characterized by the server (3) which is configured to enable recommendations to be ranked according to the preferences and features of users by using tree-based ranking algorithms.

9. A system (1) according to any one of the preceding claims; characterized by the server (3) which is configured enable the reasons for the recommendations to be explained to the users by determining the contribution of each attribute to the predictions of the model via the SHAP (Shapley Additive Explanations) technique.

10. A system (1) according to any one of the preceding claims; characterized by the server (3) which is configured to enable the recommendation list generated by the collaborative filtering technique to be customized for the user via a tree-based ranking algorithm; the decision-making processes of the algorithm to be madetransparent via the SHAP technique; and the features on the basis of which the recommendations provided to the users are selected to be provided to the users with an easily comprehensible visualization through the application (2) interface.

11. A system (1) according to any one of the preceding claims; characterized by the server (3) which is configured to enable diverse and rich content to be provided to users by integrating the collaborative filtering technique into contentbased recommendations.

Images