Personalized commodity recommendation method and system based on multi-modal data fusion

Abstract

The invention relates to a data analysis technology, and discloses a personalized commodity recommendation method and system based on multi-modal data fusion, and the method comprises the steps: recognizing registration data in account data of a user and personal information of the user, and judging whether the user is a new user or not according to the registration data; if the user is a new user, generating a random commodity list based on the personal information of the user, obtaining browsing data of the user on the random commodity list, generating a first recommended commodity according to the browsing data, and displaying the first recommended commodity to the user; otherwise, generating a second recommended commodity according to the historical purchase data of the user, identifying multi-dimensional score data of a plurality of suppliers corresponding to the second recommended commodity, generating a multi-dimensional score weight of the multi-dimensional score data according to the historical evaluation data of the user, generating a comprehensive score of each supplier according to the multi-dimensional score weight and the multi-dimensional score data, and confirming the highest-score supplier according to the comprehensive score, and recommending a second recommended commodity of the highest-score supplier to the user. The commodity recommendation accuracy can be improved.

Description

Technical Field

[0001] The present invention relates to the field of data analysis technology, and in particular to a personalized product recommendation method and system based on multimodal data fusion. Background Art

[0002] In today's digital age, the rapid development of internet technology has led to an explosive growth in product information. It's becoming increasingly difficult for consumers to sift through this vast amount of product data to find the products they truly need, leading to information overload. To effectively address this challenge, personalized product recommendation systems have emerged and have become a key technology for e-commerce platforms and other internet products to enhance user experience, increase user stickiness, and boost product sales.

[0003] Traditional personalized product recommendation methods are mainly based on single-modal data, such as making recommendations based only on users' historical purchase records, browsing behavior data, or product text descriptions. They are unable to fully utilize the value of the associations between these multi-source data, resulting in obvious limitations in the recommendation results and making it difficult to meet users' increasingly diverse and personalized needs. Summary of the invention

[0004] The present invention provides a personalized product recommendation method and system based on multimodal data fusion, the main purpose of which is to solve the problem of low accuracy in product recommendation by existing product recommendation methods.

[0005] To achieve the above objectives, the present invention provides a personalized product recommendation method based on multimodal data fusion, comprising: Obtaining user account data and identifying registration data and user personal information in the account data; Determining whether the user is a new user based on the registration data; If the user is a new user, a random product list is generated based on the user's personal information, browsing data of the user on the random product list is obtained, a first recommended product is generated based on the browsing data, and the first recommended product is displayed to the user; If the user is not a new user, obtaining the user's historical purchase data and historical evaluation data, and generating a second recommended product based on the historical purchase data; Identifying multidimensional rating data of multiple suppliers corresponding to the second recommended product, and generating multidimensional rating weights for the multidimensional rating data based on the historical evaluation data; A comprehensive score for each supplier is generated based on the multi-dimensional scoring weight and the multi-dimensional scoring data, the highest-scoring supplier is identified based on the comprehensive score, and a second recommended product of the highest-scoring supplier is recommended to the user.

[0006] Optionally, determining whether the user is a new user based on the registration data includes: Requesting the current time from a preset server to obtain the current timestamp; Extracting the registration time from the registration data to obtain the registration timestamp; Calculating the time difference between the current timestamp and the registration timestamp; Determining whether the time difference is less than a preset time difference threshold; If the time difference is less than the time difference threshold, determining that the user is a new user; If the time difference is greater than or equal to the time difference threshold, determining that the user is not a new user.

[0007] Optionally, generating a random product list based on the user's personal information includes: Constructing a user information feature map according to the user's personal information; Using a pre-trained product generation model to generate potential interest products according to the user information feature map; Generating specific product weights in combination with the current date; Dynamically weighting and sorting the potential interest products based on the specific product weights to obtain a random product list.

[0008] Optionally, generating a second recommended product according to the historical purchase data includes: Extracting the basic purchase data and implicit purchase data from the historical purchase data; Generating an initial product list according to the basic purchase data; Generating a time dynamic weight for each purchased product based on the time series; Obtaining the browsing duration, repurchase times, favorite record, and refund record of each purchased product included in the implicit purchase data; Generating a recommended weight for each purchased product according to the browsing duration, the repurchase times, the favorite record, and the refund record; Generating a recommended score for each product in the initial product list according to the time dynamic weight and the recommended weight; Confirming the second recommended product based on the recommended score.

[0009] Optionally, identifying the multi-dimensional scoring data of multiple suppliers corresponding to the second recommended product includes: Obtaining the recent evaluation data of each supplier; Extracting the direct scoring data and evaluation text data of the recent evaluation data; Normalizing the direct scoring data to obtain a normalized direct score; Generate a service type score based on the evaluation text data; Normalize the service type score to obtain a normalized service score; Fuse the normalized direct score and the normalized service score to obtain multi-dimensional score data for each supplier.

[0010] Optionally, the multi-dimensional scoring weights for generating the multi-dimensional score data based on the historical evaluation data include: Construct a keyword corpus according to the preset service categories; Identify the non-favorable evaluation data in the historical evaluation data; Extract the evaluation texts from the non-favorable evaluation data to obtain an evaluation text set; Perform sentiment intensity analysis on each evaluation text in the evaluation text set to obtain a set of sentiment intensity coefficients; Generate an attention point score for each evaluation text in the evaluation text set; Based on the set of sentiment intensity coefficients, adjust the weights of the attention point scores for each evaluation text to obtain the scoring weights for different service types in each evaluation text; Overlay the scoring weights of the same service type in all evaluation texts to obtain the multi-dimensional scoring weights.

[0011] Optionally, performing sentiment intensity analysis on each evaluation text in the evaluation text set to obtain a set of sentiment intensity coefficients includes: Perform word segmentation on the evaluation text to obtain a word segmentation result; Identify the sentiment keywords in the word segmentation result, and identify the core sentiment words and modified sentiment words of the sentiment keywords; Perform context analysis on the evaluation text to obtain a sentiment transmission path; Construct a three-dimensional sentiment grammar tree based on the core sentiment words, the modified sentiment words, and the sentiment transmission path, and calculate the sentiment base value of the sentiment grammar tree; Calculate the electroencephalogram resonance value of the sentiment grammar tree based on a pre-constructed electroencephalogram data set; Perform an analysis of the group sentiment resonance effect on the evaluation text to obtain a social sentiment resonance value; Perform weighted fusion on the sentiment base value, the electroencephalogram resonance value, and the social sentiment resonance value of each evaluation text to obtain the set of sentiment intensity coefficients.

[0012] Optionally, performing an analysis of the group sentiment resonance effect on the evaluation text to obtain a social sentiment resonance value includes: Retrieve evaluation copywriting similar to the evaluation text in the social network; Obtain the reading volume, like volume, and forwarding volume of the evaluation copywriting; Calculate the ratio of the like volume to the reading volume and multiply it by the forwarding volume to obtain the social emotional resonance value.

[0013] Optionally, generating the attention comment score for each evaluation text in the evaluation text set includes: Perform keyword matching on each evaluation text in the evaluation text set based on the keyword corpus to obtain the evaluation keyword set of each evaluation text; Identify the service type concerns of each evaluation text of the user according to the evaluation keyword set; Identify the emotional keywords of the evaluation text; Confirm the weight score of each evaluation keyword according to the emotional keyword; Generate the attention comment score of each evaluation text of the user according to the weight score and the service type concern.

[0014] To solve the above problems, the present invention also provides a personalized commodity recommendation system based on multi-modal data fusion, and the system includes: A data acquisition module for acquiring the account data of the user and identifying the registration data and user personal information in the account data; A user judgment module for judging whether the user is a new user according to the registration data. If the user is a new user, generate a random commodity list based on the user personal information, obtain the browsing data of the user for the random commodity list, generate the first recommended commodity according to the browsing data, and display the first recommended commodity to the user; A commodity generation module for judging whether the user is a new user according to the registration data. If the user is not a new user, obtain the historical purchase data and historical evaluation data of the user, and generate the second recommended commodity according to the historical purchase data; A weight analysis module for identifying the multi-dimensional scoring data of multiple supplier merchants corresponding to the second recommended commodity, and generating the multi-dimensional scoring weight of the multi-dimensional scoring data according to the historical evaluation data; A commodity recommendation module for generating the comprehensive score of each supplier merchant according to the multi-dimensional scoring weight and the multi-dimensional scoring data, confirming the highest score supplier merchant according to the comprehensive score, and recommending the second recommended commodity of the highest score supplier merchant to the user.

[0015] In an embodiment of the present invention, by obtaining the user's account data, the registration data and the user's personal information in the account data are identified. According to the registration data, it is determined whether the user is a new user. If the user is a new user, a random product list is generated based on the user's personal information, the browsing data of the user for the random product list is obtained, a first recommended product is generated according to the browsing data, and the first recommended product is displayed to the user. If the user is not a new user, the user's historical purchase data and historical evaluation data are obtained, a second recommended product is generated according to the historical purchase data, the multi-dimensional scoring weights of the multi-dimensional scoring data are generated according to the historical evaluation data, the multi-dimensional scoring data of multiple supplier merchants corresponding to the second recommended product are identified, the comprehensive score of each supplier merchant is generated according to the multi-dimensional scoring weights and the multi-dimensional scoring data, the highest score supplier merchant is confirmed according to the comprehensive score, and the second recommended product of the highest score supplier merchant is recommended to the user. Therefore, the personalized product recommendation method and system based on multi-modal data fusion proposed by the present invention can solve the problem of low accuracy in product recommendation by existing product recommendation methods. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 It is a schematic flowchart of a personalized product recommendation method based on multi-modal data fusion provided by an embodiment of the present invention; Figure 2 It is a functional module diagram of a personalized product recommendation system based on multi-modal data fusion provided by an embodiment of the present invention.

[0017] The realization, functional features and advantages of the purpose of the present invention will be further described with reference to the embodiments and the accompanying drawings. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0018] It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0019] An embodiment of the present application provides a personalized commodity recommendation method based on multimodal data fusion. The execution subject of the personalized commodity recommendation method based on multimodal data fusion includes, but is not limited to, at least one of electronic devices such as a server, a terminal, etc. that can be configured to execute the method provided in the embodiment of the present application. In other words, the personalized commodity recommendation method based on multimodal data fusion can be executed by software or hardware installed on a terminal device or a server device, and the software can be a blockchain platform. The server includes, but is not limited to: a single server, a server cluster, a cloud server, or a cloud server cluster, etc. The server can be an independent server or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, Content Delivery Network (CDN), and big data and artificial intelligence platforms.

[0020] Referring to Figure 1 As shown, it is a schematic flowchart of a personalized commodity recommendation method based on multimodal data fusion provided by an embodiment of the present invention. In this embodiment, the personalized commodity recommendation method based on multimodal data fusion includes: S1. Obtain the account data of the user and identify the registration data and user personal information in the account data.

[0021] In the embodiment of the present invention, the registration data may include the registration time of the user.

[0022] In the embodiment of the present invention, the user personal information may include information such as the age, gender, location, interest tags, consumption preferences, etc. of the user.

[0023] In the embodiment of the present invention, to identify the registration data and user personal information in the account data, the registration data and user personal information in the account data can be identified through data field semantic parsing and rule matching. First, locate the registration-related data according to preset field tags (such as "registration time", "registration IP", "account ID", etc.), and at the same time scan the content with personal identity identification characteristics such as user name, unique identity ID, mobile phone number, email, address, etc. through regular expressions and semantic models (for example, the mobile phone number needs to conform to the 11-digit number and specific number segment rules, and the unique identity ID needs to match the format of 18 digits plus a check code), and further confirm in combination with the context semantics of the field where the data is located (such as field prefixes like "user name", "personal email", etc.). Finally, complete the accurate identification through data type verification (such as the user name in the text type, the unique identity ID number field in the numerical type). During the process, it is necessary to pay attention to following privacy protection rules to desensitize sensitive information.

[0024] In the embodiments of the present invention, by obtaining the user's account data, the efficiency of identifying registration data and user personal information can be improved. By identifying the registration data and user personal information in the account data, the efficiency of determining whether the user is a new user and generating a random product list can be improved.

[0025] S2. Determine whether the user is a new user according to the registration data.

[0026] In the embodiments of the present invention, to determine whether the user is a new user according to the registration data, the current timestamp can be obtained, and the registration timestamp in the registration data can be obtained. By calculating the time difference between the two timestamps and judging the size relationship between the time difference and the preset time difference threshold, it can be determined whether the user is a new user, that is, if the time difference is less than the time difference threshold, the user is a new user, otherwise not.

[0027] In the embodiments of the present invention, determining whether the user is a new user according to the registration data includes: Request the current time from a preset server to obtain the current timestamp; Extract the registration time in the registration data to obtain the registration timestamp; Calculate the time difference between the current timestamp and the registration timestamp; Judge whether the time difference is less than a preset time difference threshold; If the time difference is less than the time difference threshold, determine that the user is a new user; If the time difference is greater than or equal to the time difference threshold, determine that the user is not a new user.

[0028] If the user is a new user, execute S3. Generate a random product list based on the user personal information, obtain the browsing data of the user for the random product list, generate the first recommended product according to the browsing data, and display the first recommended product to the user.

[0029] In the embodiments of the present invention, generating a random product list based on the user personal information is to randomly select a batch of products (such as beauty products, electronic products, clothing, etc.) from all categories of products based on the user personal information (such as age, gender, region).

[0030] In the embodiments of the present invention, generating a random product list based on the user personal information includes: Construct a user information feature map according to the user personal information; Use a pre-trained product generation model to generate potential interest products according to the user information feature map; Combine the current date to generate specific product weights; Dynamically weight and rank the potential interest products based on the specific product weights to obtain a random product list.

[0031] Specifically, constructing the user information feature map according to the user's personal information is to deeply analyze the user's personal information, map data such as age, gender, region, and interest tags into nodes, and mine the potential associations between information through semantic analysis to construct the feature map. For example, there are potential connections between young women and categories such as beauty products, fashion clothing, and online celebrity snacks. Connect these association relationships as edges to connect the nodes.

[0032] Specifically, the pre-trained product generation model is a generation model based on Transformer, which generates extended product categories related to the user's potential interests according to the user information feature map. Inputting the feature map data of the user, the model outputs product categories that the user may be interested in. For example, for users who love outdoor sports, in addition to regular sports equipment, the model may generate extended categories such as outdoor photography equipment, portable camping tables and chairs, etc.

[0033] Specifically, combining the current date to generate specific product weights means enhancing the weights for specific products by judging whether the current date is close to a holiday. For example, the weight of zongzi gift boxes will be increased when approaching the Dragon Boat Festival, and the weight of mooncake gift boxes will be increased when approaching the Mid-Autumn Festival.

[0034] In the embodiment of the present invention, when the user is a new user, by generating a random product list based on the user's personal information, obtaining the browsing data of the user for the random product list, and generating the first recommended product according to the browsing data, and displaying the first recommended product to the user, the efficiency of generating the first recommended product can be improved.

[0035] If the user is not a new user, then execute S4, obtain the user's historical purchase data and historical evaluation data, and generate the second recommended product according to the historical purchase data.

[0036] In the embodiment of the present invention, the historical purchase data includes the categories, brands, price ranges, purchase frequencies, etc. of the purchased products, and is used to analyze the user's long-term consumption habits.

[0037] In the embodiment of the present invention, the historical evaluation data refers to the evaluation data of the products purchased by the user, which can be a piece of evaluation text, or the user's satisfaction evaluation of various services, such as the score of the speed of logistics, the score of the integrity of packaging, the score of the consistency of product description, product quality, the score of customer service attitude, the score of after-sales service, etc.

[0038] In the embodiment of the present invention, generating the second recommended product according to the historical purchase data includes: Extract the basic purchase data and implicit purchase data in the historical purchase data; Generate an initial product list based on the said basic purchase data; Generate a time dynamic weight for each purchased product based on the time series; Obtain the browsing duration, repurchase times, collection records, and refund records of each purchased product included in the said implicit purchase data; Generate a recommendation weight for each purchased product based on the said browsing duration, the said repurchase times, the said collection records, and the said refund records; Generate a recommendation score for each product in the said initial product list based on the said time dynamic weight and the said recommendation weight; Confirm the second recommended product based on the said recommendation score.

[0039] Specifically, the said basic purchase data refers to structured data directly recording transaction behaviors, such as product ID, purchase time, quantity, amount, category, etc., which clearly reflect the "completed purchase behaviors" of users.

[0040] Specifically, the said implicit purchase data refers to non-direct transaction data generated during the interaction between users and products, including browsing duration, adding to cart but not purchasing, collection records, repurchase times, refund records, etc., which imply the true interest intensity and satisfaction of users with products.

[0041] Specifically, the generating of the initial product list according to the said basic purchase data may be to extract the products purchased by the user in history and their associated products (such as products of the same category, the same brand, complementary products, etc.).

[0042] Specifically, the generating of a time dynamic weight for each purchased product based on the time series means using a decay function (such as exponential decay) to assign a higher weight to products purchased recently. For example, the weight of the purchase behavior in the recent 1 month is 1, the weight 3 months ago is reduced to 0.5, and the weight 1 year ago is reduced to 0.1.

[0043] Specifically, the generating of a recommendation weight for each purchased product based on the said browsing duration, the said repurchase times, the said collection records, and the said refund records means mapping the browsing duration to a weight value between 0 and 1 through a piecewise function or normalization processing. For example, the weight is 0.8 for browsing for more than 5 minutes and 0.2 for browsing for less than 1 minute; setting a step weight according to the repurchase times, such as the weight is 1.2 for repurchasing 1 time and 1.5 for repurchasing 3 times or more; and adding a weight of 0.5 to the products collected by the user; if there is a refund record for a product, the weight of the product is reduced by 0.5 - 0.8.

[0044] In the embodiments of the present invention, by obtaining the historical purchase data and historical evaluation data of the user, the efficiency of generating the second recommended products can be improved. By generating the second recommended products according to the historical purchase data, the efficiency of identifying the multi-dimensional scoring data of multiple supplier merchants corresponding to the second recommended products can be improved subsequently.

[0045] S5. Identify the multi-dimensional scoring data of multiple supplier merchants corresponding to the second recommended products, and generate the multi-dimensional scoring weights of the multi-dimensional scoring data according to the historical evaluation data.

[0046] In the embodiments of the present invention, the multi-dimensional scoring data may include data such as the number of sold units, positive review rate, product quality score, customer service attitude score, after-sales service score, etc. of the supplier merchants.

[0047] In the embodiments of the present invention, the identification of the multi-dimensional scoring data of multiple supplier merchants corresponding to the second recommended products includes: Obtain the recent evaluation data of each supplier merchant; Extract the direct scoring data and evaluation text data of the recent evaluation data; Perform normalization processing on the direct scoring data to obtain a normalized direct score; Generate a service type score according to the evaluation text data; Perform normalization processing on the service type score to obtain a normalized service score; Fuse the normalized direct score and the normalized service score to obtain the multi-dimensional scoring data of each supplier merchant.

[0048] Specifically, the evaluation text data refers to the text of other users' evaluations of the supplier merchants.

[0049] In the embodiments of the present invention, the generation of the multi-dimensional scoring weights of the multi-dimensional scoring data according to the historical evaluation data includes: Construct a keyword corpus according to the preset service categories; Identify the non-positive review data in the historical evaluation data; Extract the evaluation text in the non-positive review data to obtain an evaluation text set; Perform sentiment intensity analysis on each evaluation text in the evaluation text set to obtain a sentiment intensity coefficient set; Generate the attention comment score of each evaluation text in the evaluation text set; Based on the sentiment intensity coefficient set, perform weight adjustment on the attention comment score of each evaluation text to obtain the scoring weights for different service types in each evaluation text; Overlay the scoring weights of the same service type in all evaluation texts to obtain the multi-dimensional scoring weights.

[0050] In the embodiment of the present invention, performing sentiment intensity analysis on each evaluation text in the evaluation text set to obtain a sentiment intensity coefficient set includes: Performing word segmentation on the evaluation text to obtain a word segmentation result; Identifying sentiment keywords in the word segmentation result, and identifying the core sentiment word and the modified sentiment word of the sentiment keyword; Performing context analysis on the evaluation text to obtain a sentiment transmission path; Constructing a three-dimensional sentiment grammar tree according to the core sentiment word, the modified sentiment word, and the sentiment transmission path, and calculating the sentiment base value of the sentiment grammar tree; Calculating the electroencephalogram resonance value of the sentiment grammar tree based on a pre-constructed electroencephalogram data set; Performing group sentiment resonance effect analysis on the evaluation text to obtain a social sentiment resonance value; Performing weighted fusion on the sentiment base value, the electroencephalogram resonance value, and the social sentiment resonance value of each evaluation text to obtain the sentiment intensity coefficient set.

[0051] Specifically, the core sentiment word can be words such as "angry", "mad", "unhappy", etc.

[0052] Specifically, the modified sentiment word can be words such as "very", "extremely", "quite", etc.

[0053] Specifically, performing context analysis on the evaluation text to obtain a sentiment transmission path means analyzing the reason for the user's sentiment, such as the causal relationship of "slow logistics - user angry".

[0054] Specifically, the electroencephalogram resonance value is a sentiment intensity index that integrates neuroscience principles, indicating the matching degree between the text sentiment expression and the human brain sentiment processing mode. When a person reads a sentiment word (such as "angry"), it will activate the corresponding brain area and generate a characteristic electroencephalogram pattern, and different electroencephalogram patterns correspond to different sentiment intensities.

[0055] Specifically, the pre-constructed electroencephalogram data set refers to the electroencephalogram records generated by preset volunteers when reading preset sentiment texts.

[0056] In the embodiment of the present invention, performing group sentiment resonance effect analysis on the evaluation text to obtain a social sentiment resonance value includes: Retrieving evaluation copywriting similar to the evaluation text in a social network; Obtaining the reading volume, like volume, and forwarding volume of the evaluation copywriting; Calculating the product of the ratio of the like volume to the reading volume and the forwarding volume to obtain the social sentiment resonance value.

[0057] Specifically, retrieving evaluation copywriting similar to the evaluation text in the social network is achieved by performing word segmentation, removing stop words, and stemming on the original evaluation text (for example, "slow logistics" and "slow express delivery" are unified as "slow logistics"), eliminating expression differences, using pre-trained models such as BERT and SBERT to convert the text into semantic vectors, and retrieving evaluations with similar semantics in the social network through cosine similarity (for example, when retrieving "extremely poor quality", expressions such as "very poor quality" and "terrible quality" are matched).

[0058] Specifically, multiplying the ratio of the number of likes to the number of reads by the number of forwards to obtain the social emotion resonance value is to measure the intensity of emotional recognition per unit exposure (to avoid the deviation of simply relying on the number of reads).

[0059] In the embodiments of the present invention, generating the attention score for each evaluation text in the evaluation text set includes: Performing keyword matching on each evaluation text in the evaluation text set based on the keyword corpus to obtain an evaluation keyword set for each evaluation text; Identifying the service type concerns of each evaluation text of the user according to the evaluation keyword set; Identifying the emotional keywords of the evaluation text; Confirming the weight score of each evaluation keyword according to the emotional keyword; Generating the attention score for each evaluation text of the user according to the weight score and the service type concerns.

[0060] Specifically, identifying the service type concerns of each evaluation text of the user according to the evaluation keyword set means confirming the service type concerns of the evaluation text according to the corresponding relationship between the keywords in the evaluation keyword set and the service type. For example, if the evaluation keyword set of the evaluation text contains keywords such as "logistics" and "quality", then it is confirmed that the service type concerns of this evaluation text are "logistics speed" and "product quality".

[0061] Specifically, confirming the weight score of each evaluation keyword according to the emotional keyword means setting the weight according to the emotional keyword of each evaluation keyword. For example, if the emotional keyword of "slow logistics speed" is "very angry", the weight score corresponding to "logistics" can be 0.1; if the emotional keyword is "extremely angry", the corresponding weight score can be 0.3; if the emotional keyword is "extremely indignant", the corresponding weight score can be 0.5.

[0062] Specifically, the weight scores corresponding to different sentiment keywords are preset and stored in the sentiment keyword data table. Even if the user's expression of sentiment keywords is different from the labeled keywords in the data table, the user's expression can be mapped to the standard expression with the closest meaning in the sentiment keyword data table through natural language processing. For example, "too angry" can be mapped to "very angry".

[0063] In the embodiment of the present invention, the step of generating the service type score according to the evaluation text data is the same as the step of generating the attention point score of each evaluation text in the evaluation text set.

[0064] S6. Generate the comprehensive score of each supplier based on the multi-dimensional scoring weight and the multi-dimensional scoring data, confirm the supplier with the highest score according to the comprehensive score, and recommend the second recommended product of the supplier with the highest score to the user.

[0065] In the embodiment of the present invention, the step of generating the comprehensive score of each supplier based on the multi-dimensional scoring weight and the multi-dimensional scoring data may be to perform normalization processing on the multi-dimensional scoring data to obtain normalized multi-dimensional scoring data, and then use the multi-dimensional scoring weight to perform weighted fusion processing on the normalized multi-dimensional scoring data to obtain the comprehensive score of each supplier.

[0066] As Figure 2 shown, it is a functional module diagram of a personalized commodity recommendation system based on multi-modal data fusion provided by an embodiment of the present invention.

[0067] The personalized commodity recommendation system 100 based on multi-modal data fusion of the present invention can be installed in an electronic device. According to the implemented functions, the personalized commodity recommendation system 100 based on multi-modal data fusion may include a data acquisition module 101, a user judgment module 102, a commodity generation module 103, a weight analysis module 104, and a commodity recommendation module 105. The modules of the present invention may also be referred to as units, which refer to a series of computer program segments that can be executed by a processor of an electronic device and can complete fixed functions, and are stored in the memory of the electronic device.

[0068] In this embodiment, the functions of each module / unit are as follows: The data acquisition module 101 is used to acquire the account data of the user and identify the registration data and the user's personal information in the account data; The user judgment module 102 is used to judge whether the user is a new user according to the registration data. If the user is a new user, a random commodity list is generated based on the user's personal information, the browsing data of the user for the random commodity list is acquired, the first recommended commodity is generated according to the browsing data, and the first recommended commodity is displayed to the user; The product generation module 103 is configured to determine whether a user is a new user according to the registration data. If the user is not a new user, the historical purchase data and historical evaluation data of the user are obtained, and a second recommended product is generated according to the historical purchase data. The weight analysis module 104 is configured to identify the multi-dimensional scoring data of multiple supplier merchants corresponding to the second recommended product, and generate the multi-dimensional scoring weights of the multi-dimensional scoring data according to the historical evaluation data. The product recommendation module 105 is configured to generate a comprehensive score for each supplier merchant according to the multi-dimensional scoring weights and the multi-dimensional scoring data, confirm the supplier merchant with the highest score according to the comprehensive score, and recommend the second recommended product of the supplier merchant with the highest score to the user.

[0069] Specifically, each module in the personalized product recommendation system 100 based on multi-modal data fusion described in the embodiments of the present invention adopts the same technical means as the Figure 1 personalized product recommendation method based on multi-modal data fusion described above, and can produce the same technical effects, which will not be elaborated here.

[0070] In the embodiments provided by the present invention, it should be understood that the disclosed devices, systems, and methods can be implemented in other ways. For example, the system embodiments described above are merely illustrative. For example, the division of the modules is only a logical function division, and there may be other division methods in actual implementation.

[0071] The modules described as separate components may or may not be physically separated. The components shown as modules may or may not be physical units, that is, they may be located in one place, or may be distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0072] In addition, each functional module in the various embodiments of the present invention may be integrated in a processing unit, or each unit may exist physically alone, or two or more units may be integrated in one unit. The above integrated units can be implemented in the form of hardware, or in the form of hardware plus software functional modules.

[0073] For those skilled in the art, it is obvious that the present invention is not limited to the details of the above exemplary embodiments, and without departing from the spirit or basic characteristics of the present invention, the present invention can be implemented in other specific forms.

[0074] Therefore, in whatever aspect, the embodiments should be regarded as exemplary and non-limiting. The scope of the present invention is defined by the appended claims rather than the above description. Thus, all changes falling within the meaning and scope of the equivalent elements of the claims are intended to be embraced by the present invention. Any reference signs in the claims should not be construed as limiting the claims concerned.

[0075] The embodiments of the present application may acquire and process relevant data based on artificial intelligence technology. Among them, artificial intelligence (AI) is a theory, method, technology, and application system that uses digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use knowledge to obtain the best results.

[0076] In addition, it is obvious that the word "comprising" does not exclude other units or steps, and the singular does not exclude the plural. The multiple units or systems stated in the system claims may also be implemented by one unit or system through software or hardware. The terms such as "first" and "second" are used to denote names and do not represent any particular order.

[0077] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not limiting. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention may be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A personalized commodity recommendation method based on multi-modal data fusion, characterized in that, The method includes: Obtain the user's account data, and identify the registration data and user personal information in the account data; Judge whether the user is a new user according to the registration data; If the user is a new user, generate a random product list based on the user personal information, obtain the browsing data of the user for the random product list, generate the first recommended product according to the browsing data, and display the first recommended product to the user; If the user is not a new user, obtain the user's historical purchase data and historical evaluation data, and generate the second recommended product according to the historical purchase data; Identify the multi-dimensional scoring data of multiple supplier merchants corresponding to the second recommended product, and generate the multi-dimensional scoring weights of the multi-dimensional scoring data according to the historical evaluation data; Generate the comprehensive score of each supplier merchant according to the multi-dimensional scoring weights and the multi-dimensional scoring data, confirm the supplier merchant with the highest score according to the comprehensive score, and recommend the second recommended product of the supplier merchant with the highest score to the user.

2. The personalized commodity recommendation method based on multimodal data fusion according to claim 1, wherein, The judging whether the user is a new user according to the registration data includes: Request the current time from a preset server to obtain the current timestamp; Extract the registration time in the registration data to obtain the registration timestamp; Calculate the time difference between the current timestamp and the registration timestamp; Judge whether the time difference is less than a preset time difference threshold; If the time difference is less than the time difference threshold, determine that the user is a new user; If the time difference is greater than or equal to the time difference threshold, determine that the user is not a new user.

3. The personalized commodity recommendation method based on multimodal data fusion according to claim 1, characterized in that, The generating a random product list based on the user personal information includes: Construct a user information feature map according to the user personal information; Use a pre-trained product generation model to generate potentially interesting products according to the user information feature map; Combine the current date to generate specific product weights; Dynamically weight and sort the potentially interesting products based on the specific product weights to obtain a random product list.

4. The personalized commodity recommendation method based on multi-modal data fusion according to claim 1, characterized in that, The generating the second recommended product according to the historical purchase data includes: Extract the basic purchase data and implicit purchase data in the historical purchase data; Generate an initial product list according to the basic purchase data; Generate the time dynamic weight of each purchased product based on the time series; Obtain the browsing duration, repurchase times, collection records, and refund records of each purchased product included in the implicit purchase data; Generate the recommendation weight of each purchased product according to the browsing duration, the repurchase times, the collection records, and the refund records; Generate the recommendation score of each product in the initial product list according to the time dynamic weight and the recommendation weight; Confirm the second recommended product based on the recommendation score.

5. The personalized commodity recommendation method based on multi-modal data fusion according to claim 1, wherein The identifying the multi-dimensional scoring data of multiple supplier merchants corresponding to the second recommended product includes: Obtain the recent evaluation data of each supplier merchant; Extract the direct scoring data and evaluation text data of the recent evaluation data; Perform normalization processing on the direct scoring data to obtain the normalized direct score; Generate the service type score according to the evaluation text data; Perform normalization processing on the service type score to obtain the normalized service score; Fuse the normalized direct score and the normalized service score to obtain the multi-dimensional scoring data of each supplier.

6. The personalized commodity recommendation method based on multi-modal data fusion according to claim 1, characterized in that The generation of the multi-dimensional scoring weights for the multi-dimensional scoring data based on the historical evaluation data includes: Construct a keyword corpus according to preset service categories; Identify the non-favorable evaluation data in the historical evaluation data; Extract the evaluation texts from the non-favorable evaluation data to obtain an evaluation text set; Perform sentiment intensity analysis on each evaluation text in the evaluation text set to obtain a sentiment intensity coefficient set; Generate the focus evaluation score for each evaluation text in the evaluation text set; Based on the sentiment intensity coefficient set, adjust the weights of the focus evaluation scores of each evaluation text to obtain the scoring weights for different service types in each evaluation text; Overlay the scoring weights of the same service type in all evaluation texts to obtain the multi-dimensional scoring weights.

7. The personalized commodity recommendation method based on multi-modal data fusion according to claim 6, wherein The performing of sentiment intensity analysis on each evaluation text in the evaluation text set to obtain a sentiment intensity coefficient set includes: Perform word segmentation on the evaluation text to obtain a word segmentation result; Identify the sentiment keywords in the word segmentation result, and identify the core sentiment words and modified sentiment words of the sentiment keywords; Perform context analysis on the evaluation text to obtain a sentiment transmission path; Construct a three-dimensional sentiment grammar tree according to the core sentiment words, the modified sentiment words, and the sentiment transmission path, and calculate the sentiment base value of the sentiment grammar tree; Calculate the electroencephalogram resonance value of the sentiment grammar tree based on a pre-constructed electroencephalogram data set; Perform group sentiment resonance effect analysis on the evaluation text to obtain a social sentiment resonance value; Perform weighted fusion on the sentiment base value, the electroencephalogram resonance value, and the social sentiment resonance value of each evaluation text to obtain the sentiment intensity coefficient set.

8. The personalized commodity recommendation method based on multimodal data fusion according to claim 7, wherein The performing of group sentiment resonance effect analysis on the evaluation text to obtain a social sentiment resonance value includes: Retrieve evaluation copywriting similar to the evaluation text in the social network; Obtain the reading volume, like volume, and repost volume of the evaluation copywriting; Multiply the ratio of the like volume to the reading volume by the repost volume to obtain the social sentiment resonance value.

9. The personalized commodity recommendation method based on multimodal data fusion according to claim 6, wherein The generation of the focus evaluation score for each evaluation text in the evaluation text set includes: Perform keyword matching on each evaluation text in the evaluation text set based on the keyword corpus to obtain an evaluation keyword set for each evaluation text; Identify the service type focus points of each user's evaluation text according to the evaluation keyword set; Identify the sentiment keywords of the evaluation text; Confirm the weight scores of each evaluation keyword according to the sentiment keywords; Generate the focus evaluation score of each evaluation text of the user according to the weight scores and the service type focus points.

10. A personalized commodity recommendation system based on multimodal data fusion, characterized in that, The system includes: A data acquisition module for acquiring the user's account data and identifying the registration data and the user's personal information in the account data; A user judgment module, which is used to judge whether a user is a new user according to the registration data. If the user is a new user, a random product list is generated based on the user's personal information, browsing data of the random product list is obtained by the user, a first recommended product is generated according to the browsing data, and the first recommended product is displayed to the user; A product generation module, which is used to judge whether a user is a new user according to the registration data. If the user is not a new user, the user's historical purchase data and historical evaluation data are obtained, and a second recommended product is generated according to the historical purchase data; A weight analysis module, which is used to identify multi-dimensional scoring data of multiple supplier merchants corresponding to the second recommended product, and generate multi-dimensional scoring weights of the multi-dimensional scoring data according to the historical evaluation data; A product recommendation module, which is used to generate a comprehensive score for each supplier merchant according to the multi-dimensional scoring weight and the multi-dimensional scoring data, confirm the supplier merchant with the highest score according to the comprehensive score, and recommend the second recommended product of the supplier merchant with the highest score to the user.

Images