Product recommendation method, system and device based on artificial intelligence

Abstract

The invention discloses a product recommendation method, system and device based on artificial intelligence, and the method comprises the steps: obtaining a historical user data set, a user demand data set and an initial product related data set, carrying out the preprocessing of the initial product related data set, and obtaining a product related data set; obtaining the preference similarity between the target user and the historical user, constructing a similar user set, and obtaining user score data in combination with the product score; obtaining the product similarity between the target product and the historical product, constructing a similar product set, and combining the product score to obtain product score data; obtaining a product feature set and a demand feature set through feature extraction, and obtaining a matched feature set through dynamic sampling and weight adjustment; and constructing a product feature matrix and a user demand vector, further obtaining feature score data, and obtaining a product recommendation result in combination with the user score data and the product score data. According to the invention, feature extraction and calculation are carried out on user and product related data, and accurate recommendation of products is realized.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and particularly relates to a product recommendation method, system and device based on artificial intelligence. Background Art

[0002] In the existing product procurement process, product screening is carried out in the following ways: users manually screen products through the product catalog given by suppliers. In this method, it is difficult for users to quickly find products that meet their needs, resulting in low procurement efficiency and a cumbersome process. Appropriate products may be missed during the manual screening process; salespersons recommend products according to the needs of users, but the accuracy and efficiency of this method are related to the experience of salespersons, and subjectivity is relatively strong in this process. The method of product recommendation by salespersons is also difficult to promote, and suppliers may lose potential users; rely on the recommendation of trading platforms, but trading platforms tend to focus more on standardized products, lack special recommendation logics, and have limited support for non-standard products. Therefore, it is difficult to meet the customized needs of users and the recommendation effect is poor.

[0003] Therefore, in the existing product recommendation methods, it is difficult for users to quickly find products that meet their own needs through manual recommendation, and this method is relatively subjective with low accuracy of product recommendation; the recommendation of trading platforms lacks a recommendation logic for products and it is difficult to meet the personalized needs of users. Summary of the Invention

[0004] The present invention aims at the deficiencies in the prior art and provides a product recommendation method, system and device based on artificial intelligence.

[0005] In order to solve the above technical problems, the present invention is solved by the following technical solutions: A product recommendation method based on artificial intelligence, comprising the following steps: Obtain a historical user data set, a user demand data set and an initial product-related data set, and preprocess the initial product-related data set based on the user demand data set to obtain a product-related data set, wherein the historical user data set includes historical users, historical products and product ratings; Based on the product ratings of the target user and historical users for historical products, obtain the preference similarity and construct a set of similar users, obtain the product ratings of the historical users in the set of similar users for the target recommended products, and obtain user rating data through the preference similarity and product ratings; Based on the product ratings of the target user and historical users for historical products and target recommended products, obtain the product similarity between the historical products and the target recommended products, construct a set of similar products, obtain the product ratings of the target user for the historical products in the set of similar products, and obtain product rating data through the product similarity and product ratings; Feature extraction is performed on the product-related data set and the user-related data set to obtain a product feature set and a demand feature set, and dynamic sampling and weight adjustment are performed on the product feature set based on the demand feature set to obtain a matching feature set; Based on the demand feature set and the matching feature set, a product feature matrix and a user demand vector are constructed, and feature scoring data is obtained through matching degree calculation. Based on the user scoring data, the product scoring data, and the feature scoring data, a product recommendation result is obtained.

[0006] As an implementable manner, obtaining the preference similarity based on the product scores of the target user and the historical users for the historical products, constructing a set of similar users, obtaining the product scores of the historical users in the set of similar users for the target recommended product, and obtaining the user scoring data through the preference similarity and the product scores include the following steps: Calculating the preference similarity between the target user and the historical users through the product scores of the target user and the historical users for the historical products, which is expressed as follows:

[0007] Constructing a set of similar users through the preference similarity, obtaining the product scores of the similar users of the target user for the target recommended product, and obtaining the user scoring data based on the preference similarity and the product scores, which is expressed as follows:

[0008] Wherein, represents the target user and the historical user the preference similarity between them, represents the target user for the historical score of the historical product, represents the target user the mean value of the historical scores, represents the historical user for the score of the historical product, represents the historical user the mean value of the historical scores, represents the target user and the historical user the set of historical products with common scores, represents the target user the user scoring data for the target recommended product, represents the historical user for the target recommended product product score, represents the target user the set of similar users.

[0009] As an implementable manner, obtaining the product similarity between the historical product and the target recommended product based on the product scores of the target user and historical users for the historical product and the target recommended product, constructing a set of similar products, obtaining the product scores of the target user for the historical products in the set of similar products, and obtaining product score data through the product similarity and the product scores, includes the following steps: Obtaining the product scores of the historical users and the target user for the historical product, and obtaining the product similarity between the historical product and the target recommended product through the product scores of the historical users for the target recommended product, which is expressed as follows:

[0010] Based on the product similarity between the historical product and the target recommended product, constructing a set of similar products for the target recommended product, and obtaining product score data through the product scores of the target product for the set of similar products and the product similarity, which is expressed as follows:

[0011] Among them, represents the historical product and the target recommended product of the product similarity, represents the target user for the th historical product score, represents the target user for the th target recommended product score of the product, represents the historical product and the target recommended product have scores of the user set, represents the product score data, represents a set of similar products.

[0012] As an implementable manner, dynamically sampling and weight-adjusting the product feature set based on the demand feature set to obtain a matching feature set, includes the following steps: Through the XGBoost pre-trained model, training based on the demand feature set and the product feature set, and inferring through the trained XGBoost model to obtain the interesting data corresponding to the product features; According to the importance distribution of the interesting data, dynamically sampling the interesting data to obtain a sampling result to form a final interesting data set; Calculating the corresponding Shapley value through the final interesting data set to obtain the first feature contribution data; Calculate the mutual information between the demand feature set and the product feature set to obtain the second feature contribution data; Analyze the interesting data of the product features through the partial dependence graph, obtain the influence relationship of the product feature set on the product feature set, and obtain the third feature contribution data; Based on the first feature contribution data, the second feature contribution data, and the third feature contribution data, obtain the feature contribution data corresponding to the product features. Based on the feature contribution data and the dynamic adjustment strategy, adjust the weights of the product feature set to obtain the matching feature set; Among them, the final interesting data set is expressed as follows:

[0013] The first feature contribution data is expressed as follows:

[0014] The second feature contribution data is expressed as follows:

[0015] Among them, represents the final interesting data obtained from the th dynamic sampling, represents the number of samples in the th dynamic sampling, represents the distribution of the th dynamic sampling, represents the importance distribution, represents the interesting data corresponding to the samples in the th dynamic sampling, represents the product feature set, represents the sample set corresponding to the final interesting data and , represents the interesting data after adding the th product feature, represents the number of features, represents the number of features in the product feature set, represents the demand feature set, represents the product feature, represents the demand feature, represents the marginal probability distribution of the product feature, represents the marginal probability distribution of the demand feature, represents the joint probability distribution.

[0016] As an implementable manner, the step of adjusting the weights of the product feature set based on the feature contribution data and the dynamic adjustment strategy to obtain the matching feature set includes the following steps: Obtain the initial weights based on the feature contribution data corresponding to the product features, and perform an initial weight adjustment on the product feature set to obtain an initial matching feature set; Based on the initial matching feature set and the initial weights, obtain the adjusted weights, and iterate the initial matching feature set to obtain the iterative matching features until the iteration condition is met to obtain the matching feature set; Among them, the initial weights and the initial matching features are expressed as follows:

[0017]

[0018] The adjusted weights and the iterative matching features are expressed as follows:

[0019]

[0020] Among them, represents the initial weights, represents the feature contribution data, represents the th matching feature, represents the th requirement feature, represents the th product feature, represents the th adjusted weight of the iteration, represents the th adjusted weight of the iteration, represents the feature range, represents the th iterative matching feature of the iteration, represents the th iterative matching feature of the iteration.

[0021] As an implementable manner, the user requirement data includes dimension requirement data, thickness requirement data, performance requirement data, delivery date requirement data, purchase price data, and category requirement data; The initial product-related data includes product dimension data, product thickness data, product performance data, product inventory data, product price data, and product category data.

[0022] As an implementable manner, the preprocessing of the initial product-related data set based on the user requirement data set to obtain the product-related data set includes the following steps: Respectively obtain the required area data and the product area data through the dimension requirement data and the product dimension data, and further obtain the area ratio data, which is expressed as follows:

[0023] An area threshold is preset, and if the area ratio data does not meet the area threshold, the corresponding product size data in the initial product-related data set is deleted to obtain a first product-related data set; The thickness difference data is obtained through the thickness requirement data and product thickness data, which is expressed as follows:

[0024] A thickness threshold is preset, and if the thickness difference data does not meet the thickness threshold, the corresponding product thickness data in the initial product-related data set is deleted to obtain a second product-related data set; The matching degree between the performance requirement data and the product performance data is obtained, and the matching degree data is obtained, which is expressed as follows:

[0025] A matching degree threshold is preset, and if the matching degree data does not meet the matching degree threshold, the corresponding product performance data in the second product-related data set is removed to obtain a third product-related data set; Obtain product demand data, combine product inventory data to obtain product delivery data, and obtain delivery score data through product delivery data and delivery demand data, as shown below:

[0026] A delivery date threshold is preset, and if the delivery date score data does not meet the delivery date threshold, the corresponding product inventory data in the third product-related data set is removed to obtain a fourth product-related data set; Eliminate product price data that does not meet the purchase price data from the fourth product-related data set, and eliminate product category data that does not meet the category requirement data, to obtain a product-related data set; in, Indicates area ratio data, Indicates the required area data, Indicates product area data, Indicates product thickness data, Indicates thickness requirement data, Indicates thickness difference data, Represents matching data, Indicates performance requirement data, Represents product demand data, Represents the delivery score data, Indicates product delivery data. Indicates the delivery demand data, Indicates a parameter.

[0027] As an implementable manner, the feature extraction of the product-related data set to obtain a product feature set and a requirement feature set includes the following steps: Obtain product light transmittance data and product strength data through product performance data, and obtain light transmittance requirement data and strength requirement data through performance requirement data; Normalize the product dimension data, product light transmittance data, product strength data, dimension requirement data, light transmittance requirement data, and strength requirement data to obtain product dimension features, product light transmittance features, initial product strength features, dimension requirement features, light transmittance requirement features, and initial requirement strength features; Perform binning processing on the product thickness data and the thickness requirement data to obtain product thickness features and requirement thickness features; Perform weight enhancement on the initial product strength feature and the initial requirement strength feature to obtain a product strength feature and a requirement strength feature, which are expressed as follows:

[0028] Among them, represents the product strength feature or the requirement strength feature, represents the weight, represents the initial product strength feature or the initial requirement strength feature; Encode the category requirement data and the product category data to obtain category requirement features and product category features; Based on the product dimension features, product light transmittance features, product thickness features, product strength features, and product category features, form a product feature set through cross-feature extraction. Based on the dimension requirement features, light transmittance requirement features, requirement thickness features, requirement strength features, and category requirement features, form a requirement feature set through cross-feature extraction.

[0029] As an implementable manner, the construction of a product feature matrix and a user requirement vector based on the requirement feature set and the matching feature set, and the calculation of feature scoring data through the matching degree include the following steps: Based on the matching feature set, construct a product feature matrix, which is expressed as follows:

[0030] Based on the requirement feature set, construct a user requirement vector, which is expressed as follows:

[0031] Calculate the similarity between the product feature matrix and the user requirement vector, and combine the matching degree weight to calculate the matching degree between the product and the user requirement to obtain feature scoring data, which is expressed as follows:

[0032] Among them, represents the product feature matrix, represents the th matching feature of the th product, represents the user demand vector, represents the th demand feature, represents the number of matching features, represents the matching degree weight of the th product feature, represents the th th product feature of the

[0033] An artificial intelligence-based product recommendation system, including a data acquisition module, a user rating module, a product rating module, a feature extraction module, and a product recommendation module; The data acquisition module acquires a historical user dataset, a user demand dataset, and an initial product-related dataset, preprocesses the initial product-related dataset based on the user demand dataset to obtain a product-related dataset, where the historical user dataset includes historical users, historical products, and product ratings; The user rating module obtains a preference similarity based on the product ratings of the target user and historical users for historical products, constructs a set of similar users, obtains the product ratings of the historical users in the set of similar users for the target recommended product, and obtains user rating data through the preference similarity and the product ratings; The product rating module obtains a product similarity between the historical product and the target recommended product based on the product ratings of the target user and historical users for historical products and the target recommended product, constructs a set of similar products, obtains the product ratings of the target user for the historical products in the set of similar products, and obtains product rating data through the product similarity and the product ratings; The feature extraction module extracts features from the product-related dataset and the user-related dataset to obtain a product feature set and a demand feature set, and dynamically samples and adjusts the weights of the product feature set based on the demand feature set to obtain a matching feature set; The product recommendation module constructs a product feature matrix and a user demand vector based on the demand feature set and the matching feature set, obtains feature rating data through matching degree calculation, and obtains a product recommendation result based on the user rating data, the product rating data, and the feature rating data.

[0034] A computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the following method is implemented: Obtain a historical user dataset, a user requirement dataset, and an initial product-related dataset, preprocess the initial product-related dataset based on the user requirement dataset to obtain a product-related dataset, where the historical user dataset includes historical users, historical products, and product ratings; Based on the product ratings of the target user and historical users for historical products, obtain a preference similarity and construct a set of similar users, obtain the product ratings of the historical users in the set of similar users for the target recommended product, and obtain user rating data through the preference similarity and product ratings; Based on the product ratings of the target user and historical users for historical products and the target recommended product, obtain the product similarity between the historical product and the target recommended product, construct a set of similar products, obtain the product ratings of the target user for the historical products in the set of similar products, and obtain product rating data through the product similarity and product ratings; Extract features from the product-related dataset and the user-related dataset to obtain a product feature set and a requirement feature set, and perform dynamic sampling and weight adjustment on the product feature set based on the requirement feature set to obtain a matching feature set; Based on the requirement feature set and the matching feature set, construct a product feature matrix and a user requirement vector, and obtain feature rating data through matching degree calculation. Based on the user rating data, product rating data, and feature rating data, obtain a product recommendation result.

[0035] A product recommendation device based on artificial intelligence includes a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, the following method is implemented: Obtain a historical user dataset, a user requirement dataset, and an initial product-related dataset, preprocess the initial product-related dataset based on the user requirement dataset to obtain a product-related dataset, where the historical user dataset includes historical users, historical products, and product ratings; Based on the product ratings of the target user and historical users for historical products, obtain a preference similarity and construct a set of similar users, obtain the product ratings of the historical users in the set of similar users for the target recommended product, and obtain user rating data through the preference similarity and product ratings; Based on the product ratings of the target user and historical users for historical products and the target recommended product, obtain the product similarity between the historical product and the target recommended product, construct a set of similar products, obtain the product ratings of the target user for the historical products in the set of similar products, and obtain product rating data through the product similarity and product ratings; Feature extraction is performed on the product-related data set and the user-related data set to obtain a product feature set and a requirement feature set, and dynamic sampling and weight adjustment are performed on the product feature set based on the requirement feature set to obtain a matching feature set; Based on the requirement feature set and the matching feature set, a product feature matrix and a user requirement vector are constructed, and feature scoring data is obtained through matching degree calculation. Based on the user scoring data, the product scoring data, and the feature scoring data, a product recommendation result is obtained.

[0036] Due to the adoption of the above technical solutions, the present invention has significant technical effects: The present invention obtains a historical user data set, a user requirement data set, and an initial product-related data set, constructs a set of similar users and a set of similar products through the historical user data set, and further obtains user scoring data and product scoring data. Feature extraction is performed on the product-related data set, and a product feature matrix and a user requirement vector are constructed based on the obtained features. Then, feature scoring data is obtained through matching degree calculation, and a product recommendation result is obtained by combining the user scoring data and the product scoring data. The method of the present invention solves the problems of low efficiency and inaccurate recommendation results in the existing method of manual screening, and supports the personalized requirements of users at the same time. Description of the Drawings

[0037] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are only some embodiments of the present invention, and those of ordinary skill in the art can obtain other drawings without creative efforts based on these drawings.

[0038] Figure 1 is a flowchart of the method of the present invention; Figure 2 is an overall schematic diagram of the system of the present invention. Detailed Embodiments

[0039] The following further describes the present invention in detail with reference to embodiments. The following embodiments are explanations of the present invention, and the present invention is not limited to the following embodiments.

[0040] A product recommendation method based on artificial intelligence, as Figure 1 shown, includes the following steps: S100. Obtain a historical user data set, a user requirement data set, and an initial product-related data set, and preprocess the initial product-related data set based on the user requirement data set to obtain a product-related data set, where the historical user data set includes historical users, historical products, and product scores; S200. Obtain the preference similarity based on the product ratings of the target user and historical users for historical products, construct a set of similar users, obtain the product ratings of the historical users in the set of similar users for the target recommended products, and obtain user rating data through the preference similarity and product ratings. S300. Based on the product ratings of the target user and historical users for historical products and target recommended products, obtain the product similarity between the historical products and the target recommended products, construct a set of similar products, obtain the product ratings of the target user for the historical products in the set of similar products, and obtain product rating data through the product similarity and product ratings. S400. Extract features from the product-related data set and the user-related data set to obtain a product feature set and a demand feature set, and perform dynamic sampling and weight adjustment on the product feature set based on the demand feature set to obtain a matching feature set. S500. Based on the demand feature set and the matching feature set, construct a product feature matrix and a user demand vector, and obtain feature rating data through matching degree calculation. Based on the user rating data, product rating data, and feature rating data, obtain the product recommendation result.

[0041] In the present invention, through the product ratings of the target user and historical users for historical products in the historical user data set, the preference similarity is obtained and a set of similar users is constructed. Through the product ratings of the historical users in the set of similar users for the target recommended products, user rating data is obtained. The product similarity between the historical products and the target recommended products is obtained, and a set of similar products is constructed, thereby obtaining product rating data. The initial product-related data set is preprocessed through the user demand data set to obtain a product-related data set. Feature extraction and weight adjustment are performed on the product-related data set, and a product feature matrix and a user demand vector are constructed. Feature rating data is obtained through matching degree calculation. Combining the user rating data and the product rating data, the product recommendation result is obtained.

[0042] In this embodiment, taking glass products as an example, a historical user data set, a user demand data set, and an initial product-related data set are obtained. Among them, the historical user data set includes historical users, historical products, and historical ratings. A historical user represents a user with a purchase behavior. A historical product represents a product purchased by the historical user in the historical purchase behavior. A historical rating represents the corresponding rating of the user for the purchased product. According to the industry characteristics of the glass base sheet, the user demand data includes size demand data, thickness demand data, performance demand data, delivery date demand data, purchase price data, and category demand data. Among them, the size demand data represents the required size of the glass base sheet by the user. Non-standard sizes are often designed in the user's requirements and need to be cut into specific specifications. The thickness demand data represents the thickness requirement of the glass base sheet by the user. In this embodiment, the performance demand data includes the light transmittance data, strength data, and processing method required by the user. The delivery date demand data refers to the user's preference for in-stock products to reduce waiting time. For non-in-stock glass, the production cycle of the supplier cannot exceed the delivery time specified by the user. The purchase price data corresponds to the user's budget limit, and the category demand data corresponds to the user's requirement for the glass category; The initial product-related data includes product size data, product thickness data, product performance data, product inventory data, product price data, and product category data. The product size data represents the standard size of the glass base sheet, such as , , the product thickness data represents the thickness range of the glass base sheet. The thickness of the glass base sheet is usually 3mm - 19mm, but for different application scenarios (such as insulating glass, tempered glass), there are specific thickness requirements. The product performance data represents the light transmittance data, strength data, and corresponding processing methods of the glass base sheet. For example, when applied to scenarios such as building facades and curtain walls, the light transmittance data of the glass base sheet needs to ensure appropriate natural lighting. The strength data measures whether the glass base sheet is suitable for tempering or laminating processing, and the processing method measures whether the glass base sheet requires bent tempered glass or other types of special-shaped glass. The product price data represents the price of each glass base sheet, and the product category data represents the types of glass base sheets, such as: clear glass, Low-E, gem blue film, Ford blue film, etc.

[0043] Preprocess the initial product-related data set based on the user demand data set to obtain the product-related data set, including the following steps: Step 1: Since the size demand data in the user demand data often involves non-standard glass base sheet sizes, the glass base sheet needs to be cut. The required area data and product area data are obtained through the size demand data and product size data respectively, and then the area ratio data is obtained, which is expressed as follows:

[0044] Step 2: In order to cut out the products required by the user with the least waste from the standard product size data, a preset area threshold is set. If the area ratio data does not meet the area threshold, in this embodiment, the area threshold is set to 30%. The corresponding product size data in the initial product-related data set is deleted to obtain the first product-related data set; Step 3: To filter out the glass substrates with appended requirement data in the data that does not meet the user's requirements, thickness difference data is obtained based on the thickness requirement data and the product thickness data, which is expressed as follows:

[0045] Step 4: The user's tolerance for thickness deviation is related to the processing and installation requirements. A preset thickness threshold is set. If the thickness difference data does not meet the thickness threshold, the corresponding product thickness data in the initial product-related data set is deleted to obtain a second product-related data set; Step 5: Regarding whether the performance meets the user's requirements, in this embodiment, the main consideration is the matching degree between the desired light transmittance of the user and the light transmittance of the original sheet. The matching degree between the performance requirement data and the product performance data is obtained to get the matching degree data, which is expressed as follows:

[0046] Step 6: A preset matching degree threshold is set. If the matching degree data does not meet the matching degree threshold, the corresponding product performance data in the second product-related data set is removed to obtain a third product-related data set; Step 7: For the product demand quantity data in the user's requirements and the product inventory data of the glass substrate, it is judged whether the delivery time of the glass substrate meets the specified time by the user. The product demand quantity data is obtained, and the product delivery date data is obtained by combining the product inventory data. Through the product delivery date data and the delivery date requirement data, the delivery date score data is obtained, which is expressed as follows:

[0047] Step 8: A preset delivery date threshold is set. If the delivery date score data does not meet the delivery date threshold, the corresponding product inventory data in the third product-related data set is removed to obtain a fourth product-related data set; Step 9: The purchase quantity of the glass substrate is usually priced by weight or area, and the discount is positively correlated with the purchase quantity. Considering the user's budget limit for the product, the product price data that does not meet the purchase price data in the fourth product-related data set is removed, and the product category data that does not meet the category requirement data is removed to obtain the product-related data set; Among them, represents the area ratio data, represents the required area data, represents the product area data, represents the product thickness data, represents the thickness requirement data, represents the thickness difference data, represents the matching degree data, represents the performance requirement data, represents the product demand data, Indicates delivery schedule scoring data, Indicates product delivery schedule data, Indicates delivery schedule requirement data, Indicates parameters.

[0048] Based on the product ratings of historical products by target users and historical users, obtain the preference similarity and construct a set of similar users, obtain the product ratings of historical users in the set of similar users for the target recommended product, and through the preference similarity and product ratings, obtain user rating data, including the following steps: Step 1: According to the ratings of the products purchased by the users, that is, the product ratings of historical products by target users and historical users, calculate the preference similarity between the target user and the historical users, which is expressed as follows:

[0049] Step 2: Through the preference similarity, construct a set of similar users. The set of similar users represents a set of several historical users with the highest preference similarity to the target user. When the target user is purchasing a product, according to the purchase situation of the set of similar users, conduct product recommendation, that is, obtain the product ratings of the similar users of the target user for the target recommended product, and based on the preference similarity and product ratings, obtain user rating data, which is expressed as follows:

[0050] Among them, Represents the target user And the historical user The preference similarity between them, Represents the target user For the first Historical rating of the historical product, Represents the target user The average value of the historical ratings, Represents the historical user For the first Rating of the historical product, Represents the historical user The average value of the historical ratings, Represents the target user And the historical user The set of historical products jointly rated, Represents the user rating data of the target user For the target recommended product, Represents the historical user For the target recommended product Product rating, Represents the target user The set of similar users.

[0051] Based on the product ratings of historical products and target recommended products by target users and historical users, obtain the product similarity between historical products and target recommended products, construct a set of similar products, obtain the product ratings of target users for historical products in the set of similar products, and through the product similarity and product ratings, obtain product rating data, including the following steps: Step 1: Obtain the product similarity by the product ratings of historical purchased products by historical users and target users, and combine with the ratings of historical users for target recommended products, that is, obtain the product ratings of historical users and target users for historical products in the historical user dataset, and obtain the product ratings of historical users for target recommended products. Based on the product ratings, obtain the product similarity between historical products and target recommended products, which is expressed as follows:

[0052] Step 2: Based on the product similarity between historical products and target recommended products, construct a set of similar products for target recommended products. The set of similar products represents a set composed of several historical products with the highest product similarity. Through the product ratings of target products for the set of similar products and the product similarity, obtain product rating data, which is expressed as follows:

[0053] Wherein, represents the historical product and the target recommended product product similarity of, represents the target user for the first rating of the historical product, represents the target user for the first product rating of the target recommended product, represents for the historical product and the target recommended product set of users with ratings for both, represents product rating data, represents the set of similar products.

[0054] Extract features from the product-related dataset and user-related dataset to obtain a product feature set and a demand feature set, including the following steps: Step 1: Obtain product light transmittance data and product strength data through product performance data, and obtain light transmittance requirement data and strength requirement data through performance requirement data; Step 2: Normalize the product size data, product light transmittance data, product strength data, size requirement data, light transmittance requirement data, and strength requirement data to obtain product size characteristics, product light transmittance characteristics, initial product strength characteristics, size requirement characteristics, light transmittance requirement characteristics, and initial requirement strength characteristics. Here, the normalization process is expressed as follows:

[0055] Wherein, represents the product size characteristics, or product light transmittance characteristics, or initial product strength characteristics, or size requirement characteristics, or light transmittance requirement characteristics, or initial requirement strength characteristics, represents the product size data, or product light transmittance data, or product strength data, or size requirement data, or light transmittance requirement data, or strength requirement data, represents the minimum value calculation function, represents the maximum value calculation function; Step 3: Perform binning on the product thickness data and thickness requirement data to obtain product thickness characteristics and required thickness characteristics. Data binning means dividing the data into several intervals and discretizing the data according to the intervals, including equal-width binning and specific binning based on business logic. For example, when the thickness range is it is binned into ; Step 4: Increase the weights for the numerical characteristics that meet the key business rules, that is, enhance the weights of the initial product strength characteristics and initial requirement strength characteristics to obtain product strength characteristics and required strength characteristics, which are expressed as follows:

[0056] Wherein, represents the product strength characteristics or required strength characteristics, represents the weight, represents the initial product strength characteristics or initial requirement strength characteristics; Step 5: Encode the category requirement data and product category data to obtain category requirement characteristics and product category characteristics. Represent the classification characteristics through embedding vectors to enhance the semantic expression ability of the characteristics, which can be used in scenarios with more processing methods or category types. At the same time, in this embodiment, feature extraction can also be performed on the processing methods. For example, the processing method is converted to , respectively corresponding to ; In this embodiment, an example of the specific feature extraction method is given as shown in Table 1 below: Table 1

[0057] Through the above-mentioned feature extraction steps, product size features, product light transmittance features, product thickness features, product strength features, product category features, size requirement features, light transmittance requirement features, required thickness features, required strength features, and category requirement features are obtained.

[0058] In this embodiment, cross-feature extraction is performed on the feature data obtained by feature extraction to form a product feature set and a requirement feature set. Through cross-feature extraction, potential relationships between multi-dimensional attributes are captured, which helps improve the accuracy and adaptability of product recommendations and reflects the combined impact of rules on multiple attribute features. The performance of the algorithm is improved through cross-features. For example: select an appropriate combination method according to industry characteristics, numerical Numerical type, generate product, weighted, or difference features, such as the product of strength and thickness; numerical Categorical type, generate segmented features to enhance the numerical correlation of categories. For example, the average light transmittance features of different categories of glass; categorical Categorical type, generate category combination features to reflect the joint distribution of multiple categories. For example, the combination of processing methods and categories; directly convert rule conditions into combination features. For example, 1 represents products that meet the condition of "light transmittance > 85 and support hot bending processing", and 0 represents those that do not. An example of cross-features is shown in Table 2 below.

[0059]

[0060] After obtaining the requirement feature set and the product feature set through cross-feature extraction and performing normalization processing, it is necessary to evaluate the importance of features. The interesting data of product features is obtained through the model and dynamically sampled, and the final interesting data set is formed based on the sampling results for evaluating the contribution of features to product recommendation performance. Comparing the impact of adding or removing certain feature data on the model performance helps select core features and adjust the weights of the product feature set based on the requirement feature set, including the following steps: Step 1: Use the model to evaluate the importance of features, such as random forest, XGBoost, LightGBM, etc. In this embodiment, through the XGBoost pre-trained model, training is performed based on the requirement feature set and the product feature set, and inference is performed through the trained XGBoost model to obtain the interesting data corresponding to the product features. Step 2: Select an importance distribution that is similar to the importance data and easy to sample as the importance distribution of the interesting data for replacing the interesting data distribution for sampling, improving the sampling efficiency and the accuracy of estimation. The importance distribution should be a distribution that is easy to generate samples, such as a uniform distribution, a normal distribution, etc. Usually, it is required that the importance distribution is non-zero on the support set of the interesting data distribution; Step 3: From the importance distribution of the interesting data Select from samples and calculate the importance weights ; Step 4: Through the importance weights of all samples, perform weighted averaging to obtain the sampling result and form the final data of interest, which is expressed as follows:

[0061] Step 5: Calculate the corresponding Shapley values through the final data of interest of different sampling results. The Shapley value analysis evaluates the importance of features by allocating the contribution of features to the prediction result, and obtains the first feature contribution data, which is expressed as follows:

[0062] Step 6: Discretize the continuous features, calculate the mutual information between the required feature set and the product feature set. The mutual information is used to measure the information sharing degree between the cross feature and the target variable, and obtain the second feature contribution data, which is expressed as follows:

[0063] Step 7: Analyze the data of interest of the product features through the partial dependence graph. Select two features as inputs, use the partial dependence graph to visualize the trend of the model output with the change of features, and the partial dependence graph measures the interaction effect of the cross feature on the model, and obtains the influence relationship of the product feature set on the product feature set, and obtains the third feature contribution data; Step 8: Based on the first feature contribution data, the second feature contribution data and the third feature contribution data, obtain the feature contribution data corresponding to the product features. The first contribution data is used to calculate the importance of a single feature, the second feature contribution data is used to measure the direct association between the feature and the target variable, and the third feature contribution data is used to evaluate the non-linear relationship between the feature and the target variable; Step 9: Through the feature contribution data and the dynamic adjustment strategy, adjust the weights of the product feature set to obtain the matching feature set, specifically: Obtain the initial weights according to the feature contribution data, and perform initial weight adjustment on the product feature set based on the initial weights to obtain the initial matching feature set, where the initial weights and the initial matching feature set are expressed as follows:

[0064]

[0065] Obtain the distance between the initial matching feature set and the required feature set, and obtain the adjusted weights based on the initial matching feature set and the initial weights; The initial matching feature set is iteratively adjusted by adjusting the weights to obtain the iterative matching features until the iterative conditions are met. In this embodiment, the iterative conditions being met means meeting the data accuracy requirement or the iterative count requirement, thus obtaining the matching feature set. Among them, the iterative matching features and the adjusted weights are expressed as follows:

[0066]

[0067] Among them, represents the final data of interest obtained from the th dynamic sampling, represents the number of samples in the th dynamic sampling, represents the distribution of the th dynamic sampling, represents the importance distribution, represents the data of interest corresponding to the samples in the th dynamic sampling, represents the product feature set, represents the sample set corresponding to the final data of interest and , represents the data of interest after adding the th product feature, represents the number of features, represents the number of features in the product feature set, represents the requirement feature set, represents the product feature, represents the requirement feature, represents the marginal probability distribution of the product feature, represents the marginal probability distribution of the requirement feature, represents the joint probability distribution, represents the initial weight, represents the feature contribution data, represents the th matching feature, represents the th requirement feature, represents the th product feature, represents the adjusted weight for the th iteration, represents the feature range, represents the iterative matching feature for the th iteration, represents the iterative matching feature for the th iteration.

[0068] For example, through analysis, it is found that the first feature contribution data of "thickness strength" is high, indicating that it makes a relatively large contribution to the prediction. The third feature contribution data shows that the influence of "light transmittance category" is a non-linear increasing relationship. The second feature contribution data of "thickness strength" is higher than other features. Then, "thickness strength" is retained as the core cross-feature and the weight is adjusted. Through this method, the initial adjustment of product features is completed.

[0069] Calculate the similarity of the cross-features in user requirements and product features, match the user requirements through glass attributes, construct a product feature matrix and a user requirement vector based on the requirement feature set and the matching feature set, and obtain the feature scoring data through the calculation of the matching degree, including the following steps: Step 1: Based on the matching feature set, construct a product feature matrix, which is expressed as follows:

[0070] Step 2: Based on the requirement feature set, construct a user requirement vector, which is expressed as follows:

[0071] Step 3: Calculate the similarity between the product feature matrix and the user requirement vector, and combine the matching degree weight to calculate the matching degree of the product and the user requirement, and obtain the feature scoring data, which is expressed as follows:

[0072] In this embodiment, a calculation example of the similarity between the product feature matrix and the user requirement vector is given: , where represents the product feature matrix, represents the th matching feature of the th product, represents the user requirement vector, represents the th requirement feature, represents the th feature scoring data of the represents the number of matching features, represents the th matching degree weight of the represents the th th product feature of the represents the product light transmittance feature in the product feature set, represents the product thickness feature, represents the light transmittance requirement feature, Represents the required thickness feature.

[0073] In this embodiment, by converting the characteristic rules of the glass industry into high-quality feature inputs, and combining dynamic weight adjustment and feature interaction processing, the business adaptability and prediction performance of the product recommendation algorithm are enhanced. This not only improves the interpretability of the model but also reflects the core role of the glass industry rules. By extracting cross features to generate new features for the glass industry, complex multi-dimensional relationships can be captured, providing richer information for the product recommendation algorithm. Combining feature normalization and dynamic weighted adjustment during the product recommendation process helps ensure the effectiveness of features and the accuracy of product recommendations.

[0074] Embodiment 2: An artificial intelligence-based product recommendation system, as Figure 2 shown, includes a data acquisition module 100, a user rating module 200, a product rating module 300, a feature extraction module 400, and a product recommendation module 500; The data acquisition module 100 acquires a historical user dataset, a user demand dataset, and an initial product-related dataset, and preprocesses the initial product-related dataset based on the user demand dataset to obtain a product-related dataset. Among them, the historical user dataset includes historical users, historical products, and product ratings; The user rating module 200 obtains the preference similarity based on the product ratings of the target user and historical users for historical products, constructs a set of similar users, obtains the product ratings of historical users in the set of similar users for the target recommended product, and obtains user rating data through the preference similarity and product ratings; The product rating module 300 obtains the product similarity between the historical product and the target recommended product based on the product ratings of the target user and historical users for historical products and the target recommended product, constructs a set of similar products, obtains the product ratings of the target user for historical products in the set of similar products, and obtains product rating data through the product similarity and product ratings; The feature extraction module 400 extracts features from the product-related dataset and the user-related dataset to obtain a product feature set and a demand feature set, and dynamically samples and adjusts the weights of the product feature set based on the demand feature set to obtain a matching feature set; The product recommendation module 500 constructs a product feature matrix and a user demand vector based on the demand feature set and the matching feature set, and obtains feature rating data through matching degree calculation. Based on the user rating data, product rating data, and feature rating data, a product recommendation result is obtained.

[0075] All changes and modifications made without departing from the spirit and scope of the present invention, and all equivalent technical solutions also fall within the scope of the present invention.

[0076] Each embodiment in this specification is described in a progressive manner. Each embodiment focuses on the differences from other embodiments. For the same or similar parts among the embodiments, reference can be made to each other.

[0077] Those skilled in the art should understand that the embodiments of the present invention can be provided as methods, devices, or computer program products. Therefore, the present invention can take the form of an all-hardware embodiment, an all-software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code.

[0078] The present invention is described with reference to the flowcharts and / or block diagrams of methods, terminal devices (systems), and computer program products according to the present invention. It should be understood that each flow and / or block in the flowchart and / or block diagram, as well as the combination of flows and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing terminal devices to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing terminal devices generate means for implementing the specified functions in one Figure 1 one flow or multiple flows and / or blocks Figure 1 or multiple blocks.

[0079] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing terminal device to work in a specific manner, so that the instructions stored in the computer-readable memory generate a manufactured article including instruction means, and the instruction means implements the specified functions in one Figure 1 one flow or multiple flows and / or blocks Figure 1 or multiple blocks.

[0080] These computer program instructions can also be loaded onto a computer or other programmable data processing terminal device, so that a series of operation steps are executed on the computer or other programmable terminal device to generate a computer-implemented process, and thus the instructions executed on the computer or other programmable terminal device provide steps for implementing the specified functions in one Figure 1 one flow or multiple flows and / or blocks Figure 1 or multiple blocks.

[0081] It should be noted that: As used in the specification, the phrase "an embodiment" or "embodiments" means that the particular features, structures, or characteristics described in connection with the embodiments are included in at least one embodiment of the present invention. Thus, the appearances of the phrase "an embodiment" or "embodiments" throughout the specification are not necessarily all referring to the same embodiment.

[0082] In addition, it should be noted that for the specific embodiments described in this specification, the shapes of the components, the names taken, etc. may be different. Any equivalent or simple changes made to the structure, features, and principles described according to the inventive concept of the present invention are included within the scope of protection of the present invention. Those skilled in the art to which the present invention pertains may make various modifications, supplements, or use similar means of substitution to the specific embodiments described, as long as they do not deviate from the structure of the present invention or exceed the scope defined by this claims, they should fall within the scope of protection of the present invention.

Claims

1. A product recommendation method based on artificial intelligence, characterized in that, It includes the following steps: Obtain the historical user dataset, user requirement dataset, and initial product-related dataset. Preprocess the initial product-related dataset based on the user requirement dataset to obtain the product-related dataset. Among them, the historical user dataset includes historical users, historical products, and product ratings; Based on the product ratings of the target user and historical users for historical products, obtain the preference similarity and construct a set of similar users. Obtain the product ratings of historical users in the set of similar users for the target recommended product. Through the preference similarity and product ratings, obtain the user rating data; Based on the product ratings of the target user and historical users for historical products and the target recommended product, obtain the product similarity between the historical product and the target recommended product, construct a set of similar products, obtain the product ratings of the target user for historical products in the set of similar products. Through the product similarity and product ratings, obtain the product rating data; Extract features from the product-related dataset and user-related dataset to obtain a product feature set and a requirement feature set, and perform dynamic sampling and weight adjustment on the product feature set based on the requirement feature set to obtain a matching feature set; Based on the requirement feature set and the matching feature set, construct a product feature matrix and a user requirement vector, and obtain feature rating data through matching degree calculation. Based on the user rating data, product rating data, and feature rating data, obtain the product recommendation result.

2. The product recommendation method based on artificial intelligence according to claim 1, wherein The step of obtaining the preference similarity and constructing a set of similar users based on the product ratings of the target user and historical users for historical products, obtaining the product ratings of historical users in the set of similar users for the target recommended product, and obtaining the user rating data through the preference similarity and product ratings includes the following steps: Calculate the preference similarity between the target user and historical users through the product ratings of the target user and historical users for historical products, as shown below: Construct a set of similar users through the preference similarity, obtain the product ratings of the similar users of the target user for the target recommended product. Based on the preference similarity and product ratings, obtain the user rating data, as shown below: in, Indicates the target user and historical users The similarity of preferences between Indicates the target user For Historical ratings of historical products, Indicates the target user The average of the historical ratings, Indicates historical users For Ratings of historical products, Indicates historical users The average of the historical ratings, Indicates the target user With historical users A collection of commonly scored historical products, Indicates the target user User rating data for target recommended products, Indicates historical users Recommend products to target Product ratings, Indicates the target user A collection of similar users.

3. The method for product recommendation based on artificial intelligence according to claim 1, characterized in that, The step of obtaining the product similarity between the historical product and the target recommended product based on the product ratings of the target user and historical users for historical products and the target recommended product, constructing a set of similar products, obtaining the product ratings of the target user for historical products in the set of similar products, and obtaining the product rating data through the product similarity and product ratings includes the following steps: Obtain the product ratings of historical users and the target user for historical products, and obtain the product similarity between the historical product and the target recommended product through the product ratings of historical users for the target recommended product, as shown below: Based on the product similarity between the historical product and the target recommended product, construct a set of similar products for the target recommended product. Through the product ratings of the target product for the set of similar products and the product similarity, obtain the product rating data, as shown below: Among them, represents the historical product and the target recommended product product similarity,[[]] represents the target user for the historical product rating,[[]] represents the target user for the target recommended product product rating,[[]] represents the historical product and the target recommended product users who have rated both,[[]] represents the product rating data,[[]] represents the set of similar products.[[]] 4. The method for product recommendation based on artificial intelligence according to claim 1, wherein The step of performing dynamic sampling and weight adjustment on the product feature set based on the requirement feature set to obtain a matching feature set includes the following steps: Through the XGBoost pre-trained model, it is trained based on the demand feature set and the product feature set, and inference is performed through the obtained XGBoost model to obtain the data of interest corresponding to the product features; According to the importance distribution of the data of interest, dynamic sampling is performed on the data of interest to obtain a sampling result to form a final data set of interest; Calculate the corresponding Shapley value through the final data set of interest to obtain the first feature contribution data; Calculate the mutual information amount between the demand feature set and the product feature set to obtain the second feature contribution data; Analyze the data of interest of the product features through the partial dependence graph, obtain the influence relationship of the product feature set on the product feature set, and obtain the third feature contribution data; Based on the first feature contribution data, the second feature contribution data, and the third feature contribution data, obtain the feature contribution data corresponding to the product features. Based on the feature contribution data and the dynamic adjustment strategy, adjust the weights of the product feature set to obtain a matching feature set; Among them, the final data set of interest is expressed as follows: The first feature contribution data is expressed as follows: The second feature contribution data is expressed as follows: Among them, represents the final data of interest obtained from the th dynamic sampling, represents the number of samples in the th dynamic sampling, represents the distribution of the th dynamic sampling, represents the importance distribution, represents the data of interest corresponding to the samples in the th dynamic sampling, represents the product feature set, represents the sample set corresponding to the final data of interest and , represents the data of interest after adding the th product feature, represents the number of features, represents the number of features in the product feature set, represents the requirement feature set, represents the product feature, represents the requirement feature, represents the marginal probability distribution of the product feature, represents the marginal probability distribution of the requirement feature, represents the joint probability distribution.

5. The product recommendation method based on artificial intelligence according to claim 4, wherein Based on the feature contribution data and the dynamic adjustment strategy, adjusting the weights of the product feature set to obtain a matching feature set includes the following steps: Obtain the initial weights according to the feature contribution data corresponding to the product features, and perform initial weight adjustment on the product feature set to obtain an initial matching feature set; Based on the initial matching feature set and the initial weights, obtain the adjusted weights, and iterate the initial matching feature set until the iteration condition is met to obtain the matching feature set; Among them, the initial weights and the initial matching features are expressed as follows: The adjusted weights and the iterative matching features are expressed as follows: Among them, represents the initial weight, represents the feature contribution data, represents the th matching feature, represents the th requirement feature, represents the th product feature, represents the th iteration's adjusted weight, represents the th iteration's adjusted weight, represents the feature range, represents the th iteration's iterative matching feature, represents the th iteration's iterative matching feature.

6. The product recommendation method based on artificial intelligence according to claim 1, wherein, The user demand data includes dimension demand data, thickness demand data, performance demand data, delivery date demand data, purchase price data, and category demand data; The initial product-related data includes product dimension data, product thickness data, product performance data, product inventory data, product price data, and product category data.

7. The method for product recommendation based on artificial intelligence according to claim 1, wherein Preprocessing the initial product-related data set based on the user demand data set to obtain a product-related data set includes the following steps: Respectively obtain the demand area data and the product area data through the dimension demand data and the product dimension data, and further obtain the area ratio data, which is expressed as follows: Preset an area threshold. If the area ratio data does not meet the area threshold, delete the corresponding product dimension data in the initial product-related data set to obtain the first product-related data set; Obtain the thickness difference data through the thickness demand data and the product thickness data, which is expressed as follows: Preset a thickness threshold. If the thickness difference data does not meet the thickness threshold, delete the corresponding product thickness data in the initial product-related data set to obtain the second product-related data set; Obtain the matching degree between the performance demand data and the product performance data to obtain the matching degree data, which is expressed as follows: A preset matching degree threshold. If the matching degree data does not meet the matching degree threshold, the corresponding product performance data in the second product-related data set is eliminated to obtain a third product-related data set; Obtain product demand data, combine product inventory data to obtain product delivery date data, and through the product delivery date data and delivery date demand data, obtain delivery date score data, which is expressed as follows: A preset delivery date threshold. If the delivery date score data does not meet the delivery date threshold, the corresponding product inventory data in the third product-related data set is eliminated to obtain a fourth product-related data set; Eliminate the product price data in the fourth product-related data set that does not meet the purchase price data, and eliminate the product category data that does not meet the category demand data to obtain a product-related data set; Among them, represents area ratio data, represents required area data, represents product area data, represents product thickness data, represents thickness requirement data, represents thickness difference data, represents matching degree data, represents performance requirement data, represents product requirement data, represents delivery time score data, represents product delivery time data, represents delivery time requirement data, represents parameters.

8. The method for product recommendation based on artificial intelligence according to claim 1, wherein Performing feature extraction on the product-related data set to obtain a product feature set and a demand feature set, including the following steps: Obtain product light transmittance data and product strength data through product performance data, and obtain light transmittance demand data and strength demand data through performance demand data; Perform normalization processing on product size data, product light transmittance data, product strength data, size demand data, light transmittance demand data, and strength demand data to obtain product size features, product light transmittance features, initial product strength features, size demand features, light transmittance demand features, and initial demand strength features; Perform binning processing on product thickness data and thickness demand data to obtain product thickness features and demand thickness features; Perform weight enhancement on the initial product strength feature and the initial demand strength feature to obtain a product strength feature and a demand strength feature, which is expressed as follows: Among them, represents the product strength characteristic or the demand strength characteristic, represents the weight, represents the initial product strength characteristic or the initial demand strength characteristic; Encode the category demand data and product category data to obtain category demand features and product category features; Based on product size features, product light transmittance features, product thickness features, product strength features, and product category features, through cross-feature extraction, form a product feature set. Based on size demand features, light transmittance demand features, demand thickness features, demand strength features, and category demand features, through cross-feature extraction, form a demand feature set.

9. The method for product recommendation based on artificial intelligence according to claim 1, wherein Based on the demand feature set and the matching feature set, construct a product feature matrix and a user demand vector, and obtain feature score data through matching degree calculation, including the following steps: Based on the matching feature set, construct a product feature matrix, which is expressed as follows: Based on the demand feature set, construct a user demand vector, which is expressed as follows: Calculate the similarity between the product feature matrix and the user demand vector, and combine the matching degree weight to calculate the matching degree between the product and the user demand to obtain feature score data, which is expressed as follows: Among them, represents the product feature matrix, represents the th matching feature of the th product, represents the user requirement vector, represents the th requirement feature, represents the feature score data of the th product, represents the number of matching features, represents the matching degree weight of the th product feature, represents the th th product feature of the 10. An artificial intelligence-based product recommendation system, characterized in that, Including a data acquisition module, a user scoring module, a product scoring module, a feature extraction module, and a product recommendation module; The data acquisition module obtains a historical user data set, a user demand data set, and an initial product-related data set, and preprocesses the initial product-related data set based on the user demand data set to obtain a product-related data set, where the historical user data set includes historical users, historical products, and product scores; The user rating module obtains the preference similarity based on the product ratings of the target user and historical users for historical products, constructs a set of similar users, obtains the product ratings of historical users in the set of similar users for the target recommended product, and obtains user rating data through the preference similarity and product ratings; The product rating module obtains the product similarity between the historical product and the target recommended product based on the product ratings of the target user and historical users for the historical product and the target recommended product, constructs a set of similar products, obtains the product ratings of the target user for the historical products in the set of similar products, and obtains product rating data through the product similarity and product ratings; The feature extraction module extracts features from the product-related data set and the user-related data set to obtain a product feature set and a demand feature set, and dynamically samples and adjusts the weights of the product feature set based on the demand feature set to obtain a matching feature set; The product recommendation module constructs a product feature matrix and a user demand vector based on the demand feature set and the matching feature set, obtains feature rating data through match degree calculation, and obtains a product recommendation result based on the user rating data, the product rating data, and the feature rating data.

11. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, the method described in any one of claims 1 to 9 is implemented.

12. An artificial intelligence-based product recommendation device, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, the method described in any one of claims 1 to 9 is implemented.

Images