An automotive product matching system and method based on large language models

Abstract

The present invention relates to the technical field of semantic understanding, and specifically, to an automotive product matching system and method based on a large language model. The method includes receiving the selling point information of a target vehicle model input by an automobile enterprise, constructing target selling point features according to the semantics of the selling point information; obtaining multiple recommended interest circles output by the large language model based on the target selling point features; obtaining the distances of different interest circles according to the matching records of multiple historical interest circles and selling point features; constructing a circle association graph of each interest circle according to the distances; mapping the multiple recommended interest circles into the circle association graph, and selecting extended interest circles within a set range around the multiple recommended interest circles; and pushing them to the user terminals of the interest circles. The present invention can match suitable people for automotive products and accurately push them to the terminals of such people, achieving the effect of precise recommendation while improving the system operation efficiency and saving network resources.

Description

Technical Field

[0001] The present invention relates to the technical field of semantic understanding, and more particularly, to a vehicle product matching system and method based on a large language model. Background Art

[0002] With the development of Internet and big data technologies, vehicle manufacturers will choose to release new vehicle models through Internet media. In the prior art, vehicle manufacturers will publish data and pictures such as descriptions and selling points of new vehicle models on public websites. Users can log in to the website to view the information of new vehicle models.

[0003] In the prior art, the way for users to obtain vehicle model information is the active viewing method, lacking the personalized recommendation function. This solution requires the vehicle model information to be published to each user's terminal without discrimination, resulting in waste of network resources and failing to achieve a good promotion effect.

[0004] In view of this, the present invention is proposed. Summary of the Invention

[0005] The purpose of the present invention is to provide a vehicle product matching system and method based on a large language model, which can match suitable people for vehicle products and accurately push them to the terminals of these people, improving the system operation efficiency, saving network resources, and achieving the effect of accurate recommendation at the same time.

[0006] To achieve the above purpose, the present invention adopts the following technical solutions:

[0007] In a first aspect, the present invention provides a vehicle product matching method based on a large language model, including:

[0008] The large language model is preliminarily trained with a first training sample, where the first training sample includes a selling point information sample and an interest circle information label of the user; the first training sample is obtained by matching according to the semantic similarity between the selling point information sample and the interest circle information label;

[0009] Construct a dynamic portrait for each of multiple interest circles;

[0010] Receive the selling point information of the target vehicle model input by the vehicle manufacturer, and construct a target selling point feature according to the semantics of the selling point information;

[0011] According to the target selling point feature, a second training sample is screened out from the data set, including a selling point feature sample and a dynamic portrait label; the data set is constructed according to the purchase behaviors of users for multiple vehicle models;

[0012] Input the target selling point features and the second training samples into the large language model, retrain the preliminarily trained large language model using the second training samples, and obtain multiple recommended interest circles output by the large language model based on the target selling point features;

[0013] According to the matching records of multiple historical interest circles and selling point features, obtain the distances of different interest circles; construct a circle association graph for each interest circle based on the distances;

[0014] Map the multiple recommended interest circles into the circle association graph, and select the extended interest circles within a set range around the multiple recommended interest circles;

[0015] Push the target vehicle model information to the user terminals of the multiple recommended interest circles and the extended interest circles through the network.

[0016] In a second aspect, the present invention provides an automobile product matching system based on a large language model, including:

[0017] A preprocessing module for preliminarily training the large language model using the first training samples, where the first training samples include selling point information samples and interest circle information labels of the users; the first training samples are obtained by matching according to the semantic similarity between the selling point information samples and the interest circle information labels; construct a dynamic portrait for each interest circle among the multiple interest circles;

[0018] An input module for receiving the selling point information of the target vehicle model input by the automobile enterprise, and constructing target selling point features according to the semantics of the selling point information;

[0019] A recommendation module for screening out the second training samples from the data set according to the target selling point features, where the data set is constructed according to the purchase behaviors of users for multiple vehicle models; the second training samples include selling point feature samples and dynamic portrait labels; input the target selling point features and the second training samples into the large language model, retrain the preliminarily trained large language model using the second training samples, and obtain multiple recommended interest circles output by the large language model based on the target selling point features; according to the matching records of multiple historical interest circles and selling point features, obtain the distances of different interest circles; construct a circle association graph for each interest circle based on the distances; map the multiple recommended interest circles into the circle association graph, and select the extended interest circles within a set range around the multiple recommended interest circles;

[0020] An output module for pushing the target vehicle model information to the user terminals of the multiple recommended interest circles and the extended interest circles through the network.

[0021] Compared with the prior art, the beneficial effects of the present invention are:

[0022] In the overall concept of the present invention, the target selling point features are input, and the recommended interest circles in the first stage are obtained by using the matching ability of the large language model. Then, based on the circle association graph, the extended interest circles in the second stage are obtained. Thus, through two-stage processing, a more comprehensive and accurate interest circle matching the target selling point features is obtained, so as to achieve the purpose of accurate recommendation and save network resources. The present invention adopts a two-stage training method for the large language model. The first training sample is constructed through language similarity, enabling the large language model to have the semantic association ability from the selling point to the interest circle. The second training sample is constructed through the dynamic portraits of the selling point features and the interest circles, enabling the large language model to deeply learn the matching relationship from the selling point to the interest circle. And when selecting the second training sample, through the contrast training method of positive and negative samples, the large language model can match the recommended interest circle closer to the target selling point features, improving the accuracy of the recommendation. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] In order to more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following will briefly introduce the drawings required for use in the description of the specific embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0024] Figure 1 is a flowchart of a method for matching automotive products based on a large language model provided by an embodiment of the present invention;

[0025] Figure 2 is a schematic structural diagram of a system for matching automotive products based on a large language model provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0026] The following describes exemplary embodiments of the present invention with reference to the accompanying drawings, including various details of the embodiments of the present invention to facilitate understanding, which should be considered merely exemplary. Therefore, those of ordinary skill in the art should recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of the present invention. Similarly, for clarity and conciseness, the description below omits the description of well-known functions and structures.

[0027] Embodiment 1

[0028] The method provided in this embodiment is applicable to the situation of pushing automotive product information to the matching user terminal according to the vehicle model selling point information (or promotional information). This method is executed by a system for matching automotive products based on a large language model. Refer to Figure 1 , and the method includes the following steps.

[0029] S110. Initially train the large language model using the first training sample.

[0030] This embodiment does not limit the type of the large language model, which can be an LLM model. The first training sample includes a selling point information sample and an interest circle information label of the user. The selling point information is text and data related to the selling points of the vehicle model. For example, equipped with a 360-degree panoramic image and a cruising range of 800 km, as a sample input into the large language model, it is called a selling point information sample. The interest circle information of the user includes the descriptive information of the interest circle. For example, the descriptive information of the self-driving circle includes: a group of people who are enthusiastic about self-driving trips, like to explore natural scenery and different cultures. The members of this circle not only enjoy driving itself, but also love planning travel routes and pursuing independent travel experiences. The interest circle information is used as a label for training the large language model, called an interest circle information label. During the actual training process, different numbers are used to represent different interest circle information labels. There is a corresponding relationship between the label and the sample, that is, a certain selling point information will attract users of certain interest circles. The corresponding relationship of the first training sample is obtained by matching according to the semantic similarity between the selling point information sample and the interest circle information label. For example, if the selling point information is suitable for self-driving and has off-road capabilities, which has a high semantic similarity with the descriptive information of the self-driving circle "enthusiastic about self-driving trips", then it constitutes the first training sample.

[0031] Perform an embedding representation on the selling point information sample to obtain a feature vector, input it into the large language model, and adjust the parameters of the large language model to make the output of the large language model approximate the interest circle information label, so as to obtain the initially trained large language model.

[0032] S120. Construct a dynamic portrait for each interest circle among multiple interest circles.

[0033] The dynamic portrait D of each interest circle represents the characteristics of the interest circle, and as the focus or interest point of the users within the interest circle changes, the dynamic portrait of the interest circle changes dynamically over time.

[0034] S130. Receive the selling point information of the target vehicle model input by the vehicle enterprise, and construct target selling point features according to the semantics of the selling point information.

[0035] The vehicle enterprise inputs the selling point information of a target vehicle model through the terminal, such as large space and energy saving. In this embodiment, semantic extraction and analysis are performed on the selling point information to form a feature vector, called a target selling point feature.

[0036] S140. Screen out the second training sample from the dataset according to the target selling point feature and the dynamic portrait.

[0037] The large language model after preliminary training only has the function of matching interest circles according to descriptive selling point information. The second training sample (including selling point feature samples and dynamic portrait labels) is used to retrain the large language model after preliminary training. By performing semantic extraction and analysis on the selling point information, the selling point features can better reflect the characteristics of the selling points; by constructing a dynamic portrait from the descriptive interest circle information, the characteristics of the interest circles can also be better reflected. Input the selling point feature samples into the large language model and continue to adjust the parameters of the large language model so that the output of the large language model approaches the dynamic portrait labels. There is a one-to-one correspondence between the dynamic portrait labels and the interest circles. Therefore, the trained large language model can be used to obtain the interest circles corresponding to the target selling point features.

[0038] The process of obtaining the second training sample is described below: First, construct a dataset based on the purchase behaviors of users for various vehicle models. For example, if the interest circles to which users who purchased Model M belong are the self-driving circle and the anime circle, then the selling point feature samples of Model M and the dynamic portrait labels of the self-driving circle and the anime circle form the second training sample. Since the dataset stores the selling point features of various vehicle models and the features of multiple dynamic portraits, and the target selling point feature is the feature of a certain vehicle model, in order to enable the large language model to have a more accurate matching ability for this vehicle model, it is necessary to screen out the second training samples (including positive samples and negative samples) from the dataset that have a certain semantic similarity with the target selling point feature. Specifically, calculate the semantic similarity between the target selling point feature and each selling point feature in the dataset, and use the selling point feature samples with a semantic similarity greater than the high threshold and the corresponding dynamic portrait labels as positive samples, and use the selling point feature samples with a semantic similarity less than the low threshold and the corresponding dynamic portrait labels as negative samples. Among them, the high threshold is 90% and the low threshold is 10%. Assume that the target selling point feature is represented by T 4 and use BERT to perform semantic representation on T 4 to obtain the feature vector of the selling point , see formula (1). The selling point features in the dataset are also processed by BERT into feature vectors . Use cosine similarity ( ) to calculate the similarity between T 4 and the selling point feature vectors in the dataset, see formula (2). Use the selling point features with a similarity greater than 90% and the corresponding dynamic portrait labels as positive samples, and use the selling point features with a similarity less than 10% and the corresponding dynamic portrait labels as negative samples.

[0039] ; (1)

[0040] ; (2)

[0041] S150. Input the target selling point features and the second training samples into the large language model, retrain the preliminarily trained large language model using the second training samples, and obtain multiple recommended interest circles output by the large language model based on the target selling point features.

[0042] Specifically, fill the target selling point features into the question part of the prompt template, fill the second training samples into the example part of the prompt template, and fill the recommended interest circles into the answer part of the prompt template. For example, the prompt template includes: The selling points of the vehicle models that users are interested in can be summarized as {T 4}. Please perform a correlation analysis based on the dynamic portraits {D} of different interest circles and predict which interest circles are the audiences for this selling point. The following are some examples: {positive samples, negative samples}. Among them, {} are the parts that need to be filled.

[0043] Input the filled prompt template into the large language model. It is pre-agreed that the examples in the prompt template are taken out as the second training samples. The language model is retrained using the second training samples. After training is completed, extract the target selling point features T 4 as the input of the large language model. The large language model obtains multiple dynamic portraits based on the target selling point features, and then obtains the interest circles corresponding to the dynamic portraits, which are called recommended interest circles. Exemplarily, call the large language model to obtain the TOP-K recommended interest circles, and K can be 3.

[0044] S160. Obtain the distances of different interest circles according to the matching records of multiple historical interest circles and selling point features; construct a circle association graph for each interest circle based on the distances.

[0045] This step mainly constructs a circle association graph based on the distances of different interest circles for subsequent expansion of the recommended interest circles. Since the dataset stores the selling point features of multiple vehicle models and the features of multiple dynamic portraits, in this embodiment, obtain the matching records of multiple historical interest circles and selling point features from the aforementioned dataset. First, for the aforementioned matching records, merge the semantically similar selling point features. It can be understood that when the selling point features are merged, the corresponding interest circles will also be merged. Count the number of times that different interest circles are recommended together (that is, the number of times they appear in the recommended circles of the same selling point feature), and the similarity of the static portraits S of different interest circles; obtain the distances of different interest circles according to the number of times and the similarity. Calculate the distances of different interest circles using formula (3) and formula (4) .

[0046] ; (3)

[0047] ; (4)

[0048] Among them, and are respectively the two static portraits of two interest circles, is the result obtained by subtracting the similarity of two interest circles from 1, is the interest circle co-occurrence times, is a fixed hyperparameter.

[0049] Finally, an interest circle association graph is obtained, where represents the nodes of the interest circles, represents the distance between the interest circles.

[0050] S170. Map multiple recommended interest circles into the interest circle association graph, and select the extended interest circles within the set range around the multiple recommended interest circles.

[0051] Determine the nodes consistent with each recommended interest circle in the interest circle association graph; take any node as the center, and traverse the nodes within the set range around it as the extended interest circles. Suppose the self-driving circle is a recommended interest circle, then in the interest circle association graph, take the self-driving circle node as the center, draw a circle with the set range (for example, 5), and take the nodes within the circle that are directly connected / indirectly connected to the self-driving circle node as the extended interest circles.

[0052] S180. Push the target vehicle model information to the user terminals of multiple recommended interest circles and extended interest circles through the network.

[0053] The target vehicle model information includes the selling points, vehicle configuration, pictures, etc. of the target vehicle model. The relationship between the user and the interest circle can be pre-constructed. For example, take the social groups on the self-driving website as the users of the self-driving circle.

[0054] Optionally, preset the number of interest circles to be promoted, for example, 3. If the sum of the number of recommended interest circles and extended interest circles is less than or equal to 3, directly push the target vehicle model information to the user terminals of multiple recommended interest circles and extended interest circles. If the sum of the number of recommended interest circles and extended interest circles is greater than 3, it is necessary to obtain 3 interest circles through the weighted sampling algorithm. For example, according to the historical push records of the user terminals, perform weighted sampling on the extended interest circles and recommended interest circles.

[0055] The following formula is used to sample the circles:

[0056] ; (5)

[0057] ; (6)

[0058] Among them, is the sampling weight of the i-th interest circle layer, represents the i-th interest circle layer, is a fixed hyperparameter, is the number of times of pushing vehicle model information to the user terminals in the interest circle layer . The more times of pushing, the lower the sampling weight, which is beneficial to improving the diversification of the pushed interest circle layers. represents the weighted sampling algorithm, represents the sampling weights of each interest circle layer, represents the set of recommended interest circle layers and extended interest circle layers, represents the interest circle layer returned by sampling.

[0059] In the overall idea of the present invention, by inputting the target selling point features, the recommended interest circle layers in the first stage are obtained by using the matching ability of the large language model, and then based on the circle layer association graph, the extended interest circle layers in the second stage are obtained, so as to obtain more comprehensive and accurate interest circle layers that match the target selling point features through two-stage processing, thereby achieving the purpose of accurate recommendation and saving network resources. The present invention adopts a two-stage training method for the large language model, namely preliminary training and re-training. The first training sample is constructed through language similarity, enabling the large language model to have the semantic association ability from the selling points to the interest circle layers. The second training sample is constructed through the dynamic portraits of the selling point features and the interest circle layers, enabling the large language model to deeply learn the matching relationship from the selling points to the interest circle layers. And when selecting the second training sample, through the contrast training method of positive and negative samples, the large language model can match the recommended interest circle layers closer to the target selling point features, improving the accuracy of the recommendation.

[0060] Since there is a strong correlation between vehicle model selling points (automobile brand marketing) and some circle layers, such as the self-driving circle, cycling circle, modification circle, etc., it is difficult to avoid the problem of homogenization of recommendation results in the prompt engineering on the large language model, that is, given different descriptions of vehicle model selling points, the model will recommend similar circle layers and lacks the ability to explore the correlation between different circle layers. Therefore, a two-stage interest circle layer recommendation method is proposed. Based on the prompt engineering of the large language model, the correlation analysis of different circle layers is carried out for a large amount of recommendation data generated in the actual production scenario, improving the diversity and comprehensiveness of the model recommendation results.

[0061] Embodiment 2

[0062] Based on the above embodiment, this embodiment optimizes the process of obtaining the dynamic portrait. Constructing the dynamic portrait of each interest circle layer in multiple interest circle layers includes the following three steps:

[0063] Step 1: Obtain the user information I of multiple users. Identify the interest circles to which the users belong and the static portraits S of the interest circles based on the user information I. The user information includes information such as age, region, interests, consumption, and media touchpoints (active social media).

[0064] The portrait modeling model is, for example, an LLM model. Use the user information to construct a prompt template , and input the prompt template into the portrait modeling model (such as an LLM model) to obtain the static portrait output by the portrait modeling model. The prompt template is, for example: "The user information is {I}. Please perform portrait modeling on this circle, analyze user behavior from the perspective of sociology, etc., and speculate on the possible needs of the circle group for the characteristics of car models based on their behavior. Here are some examples: {}" The variables within {} can be automatically replaced. The examples are as follows:

[0065] The self-driving circle refers to a group of people who are enthusiastic about self-driving travel and like to explore natural scenery and exotic cultures. Members of this circle not only enjoy driving itself but also love planning travel routes and pursuing independent travel experiences. The static portrait of the self-driving circle includes: The self-driving circle is mainly composed of young and middle-aged people and family groups, with a balanced gender distribution. They usually have a certain amount of economic strength and free time. Members of the self-driving circle prefer free and flexible travel methods and enjoy the exploration process on the road, especially through diverse travel locations such as natural scenery and cultural sites. This group likes to share experiences such as self-driving routes, journey scenery, and travel equipment on social platforms. The consumption focus of self-driving circle users is on high-quality, reliable, and comfortable travel equipment, such as camping equipment, in-car refrigerators, portable chargers, etc. The vehicle model requirements of the self-driving circle are also part of the static portrait. The self-driving circle group attaches particular importance to the high passability, comfort, and long-distance driving performance of vehicles, and especially favors models suitable for various road conditions such as SUVs or off-road vehicles. Large space and flexible storage design are important requirements to meet the needs of storing equipment during self-driving travel. This group pays attention to the reliability, safety, and fuel economy of vehicles and prefers safety configurations such as electronic stability control systems, panoramic cameras, and autonomous driving assistance. In addition, self-driving circle members also have relatively high requirements for configurations such as in-car navigation systems, long-distance driving assistance, and seat comfort to enhance the comfort and convenience of long-distance journeys.

[0066] Step 2: Obtain the hot topic data of the interest circles.

[0067] Specifically, relevant data is scraped from social media platforms (data is crawled according to catalysts), and natural language processing technology is used for data cleaning, text tokenization, and word frequency statistics to identify the hot topic data M of each interest circle. Among them, data cleaning is to preprocess the original data and remove irrelevant information (such as punctuation marks, emojis, extra spaces, advertising information, etc.). This method uses regular expressions (Regex) to remove URLs, HTML tags, special characters, and filter out stop words (such as "de", "shi", etc.) to reduce the interference of irrelevant vocabulary. Text tokenization is to divide continuous text into several words or phrases. This method uses the Jieba tokenization tool, and its core principle is: based on the DAG (directed acyclic graph) search of the prefix dictionary, and combined with the dynamic programming algorithm to find the maximum probability path, so as to achieve the optimal tokenization result. Word frequency statistics is to count the tokenized text to identify the words with higher frequencies. This method uses the TF-IDF (term frequency-inverse document frequency) algorithm to implement.

[0068] Step 3: Use the static portrait S of the interest circle and the hot topic data M to construct a prompt template, and input the prompt template into the portrait modeling model to obtain the dynamic portrait output by the portrait modeling model. The dynamic portrait includes the vehicle model requirements of the user.

[0069] Construct a prompt template It is required that the portrait modeling model can be based on the static portrait and the hot topic data M to analyze the interest hotspots and vehicle model requirements of different circles, so as to perform periodic dynamic updates on the static portrait based on the social media hot data.

[0070] The core idea of constructing the prompt template is as follows: "The user information is { }, and in the past period of time, the hot topics that the users in this circle have paid attention to on social media are { }. Please perform portrait modeling on this circle, analyze the user behavior from the perspective of sociology, etc., summarize its interest hotspots, and infer the possible requirements of this circle group for the characteristics of car models according to their behavior."

[0071] According to the actual input data and , format the prompt template and call the portrait modeling model to perform portrait modeling to obtain the dynamic portrait of the circle .

[0072] This embodiment has the following technical effects:

[0073] There is a gap between existing portrait modeling methods and actual production scenarios. The features obtained from modeling are often too general and lack relevance to specific application requirements. For example, there are differences between the information such as the age distribution and geographical distribution of users obtained from modeling analysis and the interest distribution related to the selling points of vehicle models required in the actual application scenario. To address this issue, this embodiment proposes a circle portrait modeling method for vehicle model demand analysis to ensure high relevance and high availability between the features after modeling and actual production requirements.

[0074] Over time, the focus and interests of circles may shift. To better adapt to the dynamic changes in circle characteristics, it is necessary to perform dynamic modeling of circles periodically. To address this issue, this embodiment proposes a method for periodically updating the static circle portrait based on social media hot data, thereby effectively capturing the interest changes of the current circle.

[0075] Embodiment 3

[0076] Based on the above embodiment, this embodiment optimizes the construction process of target selling point features. According to the semantics of the selling point information, target selling point features are constructed, including the following operations:

[0077] The first step: Perform syntactic parsing and semantic analysis on the selling point information T 1 to obtain a selling point-description binary tuple.

[0078] Extract the selling point-description binary tuple T 1 from the selling point information T 2 , and merge the binary tuples with similar semantics.

[0079] The second step: Use the automotive knowledge base to perform proper noun detection and explanation on the binary tuple to form a selling point-description-explanation triple.

[0080] Based on the existing automotive proper noun list, use the RAG (Retrieval-Augmented Generation) architecture to explain proper nouns in the automotive field. If the proper noun exceeds the local knowledge scope, call the search tool to obtain the corresponding noun explanation from the network, and finally construct the automotive knowledge base. Perform proper noun detection on the selling point-description binary tuple. If the binary tuple involves proper nouns in the automotive field, perform noun explanation through the automotive knowledge base to obtain the selling point-description-explanation triple T 3 .

[0081] The third step: Perform correlation analysis on the triple and automotive brand marketing to form a selling point-description-explanation-correlation quadruple, and retain the quadruples with a correlation greater than the set threshold as the target selling point features.

[0082] Perform a correlation analysis on the selling point-description-explanation triple and automotive brand marketing, that is, determine whether the triple belongs to the category of automotive brand marketing and obtain a correlation score. For example, determine the probability that the triple belongs to automotive brand marketing through a classification model, and use this probability as the correlation score. This classification model is pre-trained using multiple triple samples belonging to automotive brand marketing. Remove triples with a correlation less than or equal to a set threshold (e.g., 70%), retain triples with a correlation greater than the set threshold, and append the specific correlation score at the end to form a selling point-description-explanation-correlation quadruple T 4 。

[0083] This embodiment has the following technical effects:

[0084] Due to the uncertainty of the selling point information input by the user, to eliminate the interference of issues such as irrelevance to automotive brand marketing, repeated emphasis on single selling points, and grammatical redundancy in the selling point information, for this problem, a text description normalization method for vehicle model selling points is proposed to parse and normalize the selling point information input by the user, reducing performance losses caused by the uncertainty of the selling point information input by the user.

[0085] Embodiment 3

[0086] This embodiment provides an automotive product matching system based on a large language model. See Figure 2 ,including:

[0087] A preprocessing module for preliminarily training the large language model using a first training sample, where the first training sample includes selling point information samples and interest circle information tags of the user; the first training sample is obtained by matching according to the semantic similarity between the selling point information samples and the interest circle information tags; constructing a dynamic portrait of each interest circle in multiple interest circles;

[0088] An input module for receiving the selling point information of the target vehicle model input by the vehicle enterprise and constructing target selling point features according to the semantics of the selling point information;

[0089] A recommendation module for screening out second training samples from a dataset according to the target selling point features, including selling point feature samples and dynamic portrait labels; the dataset is constructed based on the purchase behaviors of users for various vehicle models; inputting the target selling point features and the second training samples into a large language model, retraining the preliminarily trained large language model with the second training samples, and obtaining multiple recommended interest circles output by the large language model based on the target selling point features; obtaining the distances of different interest circles according to the matching records between historical multiple interest circles and selling point features; constructing a circle association graph for each interest circle according to the distances; mapping the multiple recommended interest circles into the circle association graph, and selecting extended interest circles within a set range around the multiple recommended interest circles;

[0090] An output module for pushing the target vehicle model information to the user terminals of multiple recommended interest circles and extended interest circles through the network.

[0091] The system provided in this embodiment can execute the methods provided in Embodiments 1 to 3 and has corresponding technical effects.

[0092] It should be understood that various forms of the processes shown above can be used, reordering, adding or deleting steps. For example, the steps described in the present invention can be executed in parallel, sequentially or in a different order, as long as the desired results of the technical solutions disclosed in the present invention can be achieved, and no limitations are imposed herein.

[0093] The above specific embodiments do not constitute a limitation to the protection scope of the present invention. Those skilled in the art should understand that various modifications, combinations, sub - combinations and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions and improvements made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A method for matching automobile products based on a large language model, characterized in that: include: The large language model is preliminarily trained using a first training sample, where the first training sample includes a selling point information sample and an interest circle information label to which the user belongs; The first training sample is obtained by matching the selling point information sample with the semantic similarity of the interest circle information label; Build a dynamic profile of each interest circle in multiple interest circles; Receive selling point information of a target vehicle model input by a car manufacturer, and construct target selling point features according to the semantics of the selling point information; According to the target selling point feature, a second training sample is selected from the data set, including a selling point feature sample and a dynamic portrait label; the data set is constructed according to the user's purchase behavior of multiple car models; Inputting the target selling point feature and the second training sample into a large language model, retraining the large language model that has been preliminarily trained using the second training sample, and obtaining a plurality of recommended interest circles output by the large language model based on the target selling point feature; According to the matching records of multiple interest circles and selling point features in history, the distances between different interest circles are obtained; and according to the distances, a circle association graph of each interest circle is constructed; Mapping the multiple recommended interest circles into a circle association graph, and selecting an extended interest circle within a set range around the multiple recommended interest circles; Pushing the target vehicle model information to user terminals of multiple recommended interest circles and extended interest circles through the network; According to the target selling point feature, a second training sample is screened out from the data set, including: calculating the semantic similarity between the target selling point feature and each selling point feature in the data set, taking the selling point feature samples and the corresponding dynamic portrait labels whose semantic similarity is greater than a high threshold as positive samples, and taking the selling point feature samples and the corresponding dynamic portrait labels whose semantic similarity is less than a low threshold as negative samples.

2. The automobile product matching method based on a large language model according to claim 1, characterized in that: Build dynamic portraits of each interest circle in multiple interest circles, including: Obtain user information of multiple users, identify interest circles to which the users belong based on the user information, and static portraits of the interest circles; Obtain hot topic data in interest circles; According to the static portrait of the interest circle and the hot topic data, a dynamic portrait of the interest circle is obtained.

3. The automobile product matching method based on a large language model according to claim 2 is characterized in that: Identify the interest circle to which the user belongs according to the user information, and the static portrait of the interest circle includes: Using the user information to construct a prompt template, inputting the prompt template into a portrait modeling model, and obtaining a static portrait output by the portrait modeling model; According to the static portrait of the interest circle and the hot topic data, a dynamic portrait of the interest circle is obtained, including: The static portrait of the interest circle and the hot topic data are used to construct a prompt template, and the prompt template is input into the portrait modeling model to obtain a dynamic portrait output by the portrait modeling model.

4. The automobile product matching method based on a large language model according to claim 1, characterized in that: Constructing target selling point features according to the semantics of the selling point information, including: Performing syntax parsing and semantic analysis on the selling point information to obtain a selling point-description tuple; Using the automobile knowledge base to perform proper noun detection and interpretation on the bigrams to form selling point-description-explanation triplets; A correlation analysis is performed on the triples and automobile brand marketing to form a selling point-description-explanation-correlation quadruple, and the quadruple with a correlation greater than a set threshold is retained as the target selling point feature.

5. The automobile product matching method based on a large language model according to claim 1, characterized in that: The target selling point feature and the second training sample are input into the large language model, the large language model that has been preliminarily trained is retrained using the second training sample, and multiple recommended interest circles output by the large language model are obtained based on the target selling point feature, including: Fill the target selling point feature into the question part of the prompt template, fill the second training sample into the example part of the prompt template, and fill the recommended interest circle into the answer part of the prompt template; The filled prompt template is input into the large language model so that the large language model is retrained with the second training sample, and a plurality of recommended interest circles output by the large language model are obtained based on the target selling point features.

6. The automobile product matching method based on a large language model according to claim 1, characterized in that: Based on the historical matching records of multiple interest circles and selling point features, the distances between different interest circles are obtained, including: Merge semantically similar selling point features; Count the number of times different interest circles recommend something together, and the similarity of static portraits of different interest circles; The distances between different interest circles are obtained according to the number and similarity.

7. The automobile product matching method based on a large language model according to claim 6, characterized in that: Mapping the multiple recommended interest circles to a circle association graph, and selecting an extended interest circle within a set range around the multiple recommended interest circles, including: Determine the nodes consistent with each recommended interest circle in the circle association graph; Taking any node as the center, traverse the nodes within the surrounding set range as the expanded interest circle.

8. The automobile product matching method based on a large language model according to claim 7 is characterized in that: After traversing the nodes within the surrounding set range as the extended interest circle, it also includes: According to the historical push records to the user terminal, weighted sampling is performed on the extended interest circle and the recommended interest circle.

9. An automobile product matching system based on a large language model, characterized in that: include: A preprocessing module, configured to perform preliminary training on the large language model using a first training sample, wherein the first training sample includes a selling point information sample and an interest circle information label to which the user belongs; The first training sample is obtained by matching the selling point information sample with the semantic similarity of the interest circle information label; constructing a dynamic portrait of each interest circle in the multiple interest circles; An input module is used to receive the selling point information of the target vehicle model input by the car manufacturer, and construct the target selling point features according to the semantics of the selling point information; A recommendation module, for selecting a second training sample from a data set according to the target selling point feature, including a selling point feature sample and a dynamic portrait label; the data set is constructed according to the user's purchase behavior for a variety of car models; the target selling point feature and the second training sample are input into a large language model, the large language model that has been preliminarily trained is retrained using the second training sample, and multiple recommended interest circles output by the large language model are obtained based on the target selling point feature; the distances between different interest circles are obtained according to the historical matching records of multiple interest circles and selling point features; a circle association graph of each interest circle is constructed according to the distance; the multiple recommended interest circles are mapped into the circle association graph, and extended interest circles within a set range around the multiple recommended interest circles are selected; wherein, selecting a second training sample from a data set according to the target selling point feature includes: calculating the semantic similarity between the target selling point feature and each selling point feature in the data set, taking the selling point feature samples and the corresponding dynamic portrait labels whose semantic similarity is greater than a high threshold as positive samples, and taking the selling point feature samples and the corresponding dynamic portrait labels whose semantic similarity is less than a low threshold as negative samples; The output module is used to push the target vehicle model information to user terminals of multiple recommended interest circles and extended interest circles through the network.

Images