A product recommendation method and system based on state update

Abstract

The present application relates to the field of intelligent recommendation. In particular, it relates to a product recommendation method and system based on state updating. The method comprises obtaining product state time series data and consumer preference state time series data, inputting them into a recommendation model based on a double-state dynamic neural network, outputting product feature vectors and consumer preference feature vectors, and obtaining a pre-sale product recommendation list using a collaborative filtering method. According to the pre-sale product recommendation list, a corresponding adoption product is selected. After the product growth process is completed, the product state time series data and the consumer preference state time series data of the product that has not been adopted are input into the recommendation model based on the double-state dynamic neural network, real-time product feature vectors and real-time consumer preference feature vectors are output, and a formal product recommendation list is obtained using a collaborative filtering method. The present application can take into account the whole process of agricultural products and dynamically adjust the recommendation content according to the state of agricultural products.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of intelligent adoption and intelligent recommendation of agricultural products. In particular, it relates to a product recommendation method and system based on state updating. BACKGROUND

[0002] With the growth of China's economy and the improvement of people's living standards, people pay more and more attention to healthy diet, green food and food safety, etc., so the quality requirements for agricultural products are also getting higher and higher. In order to meet the needs of consumers, agricultural producers and suppliers need to continuously improve the quality and safety of products to meet market demand. As a result, in recent years, a new consumption mode has emerged, that is, consumers adopt agricultural products to participate in the process of agricultural production and understand the growth status and breeding process of livestock or aquatic products in real time. Through the way of consumer adoption, the added value and market competitiveness of agricultural products can be improved, and at the same time, the trust and satisfaction of consumers to agricultural products can be promoted, and the awareness of consumers to food safety can be improved. The recommendation system refers to a system that uses machine learning, artificial intelligence and other technologies to analyze the historical behavior and preferences of consumers, and then recommends the most suitable goods or services to them.

[0003] Chinese patent CN114066482A discloses a food traceability intelligent recommendation system and method, which considers traceability information in the recommendation system and takes the traceability information of the origin and production process of food as static information into consideration of commodity characteristics. For example, Chinese patent CN114819387A discloses a time sequence recommendation method and system based on deep reinforcement learning, which uses a time sequence recommendation system to update the state of consumers.

[0004] In the above-mentioned recommendation scheme, the breeding process information and real-time situation of the product are not fully considered, and the real-time state of the product is not considered, which cannot provide the most interesting adopted agricultural products or agricultural product commodities for consumers. SUMMARY

[0005] Based on the problems existing in the prior art, the present application proposes a product recommendation method and system based on state updating, which can better meet the needs of consumer adoption and serve agricultural product sales. First, compared with traditional recommendation methods, the present application will comprehensively consider the breeding process information and real-time situation of the product, dynamically recommend the adopted product, and provide more suitable choices for consumer adoption. Second, the present application can help breeders better sell products and help agricultural product sales. Third, the present application can help breeders improve production efficiency and management level. Through the present application, breeders can better understand the needs and preferences of consumers, and carry out targeted breeding and management to improve production efficiency and management level, and reduce production cost and risk.

[0006] In order to achieve the above object, the present application provides the following technical solutions.

[0007] In the first aspect of the present application, the present application provides a product recommendation method based on state update, comprising:

[0008] 101. Obtain the historical state and the current state of the product growth process to form product state time series data;

[0009] 102. Obtain the historical preference state data and the current preference state data of the consumer consumption process to form consumer preference state time series data;

[0010] 103. Input the product state time series data and the consumer preference state time series data into a recommendation model based on a double-state dynamic neural network to output real-time product feature vectors and real-time consumer preference feature vectors;

[0011] 104. According to the real-time product feature vectors and real-time consumer preference feature vectors output by step 103, a pre-sale product recommendation list is obtained by using a collaborative filtering method;

[0012] 105. According to the pre-sale product recommendation list, a corresponding adoption product is selected;

[0013] 106. Input the product state time series data and the consumer preference state time series data of the product which is not adopted after the product growth process into a recommendation model based on a double-state dynamic neural network to output real-time product feature vectors and real-time consumer preference feature vectors;

[0014] 107. According to the real-time product feature vectors and real-time consumer preference feature vectors output by step 106, a formal product recommendation list is obtained by using a collaborative filtering method.

[0015] In the second aspect of the present application, the present application provides a product recommendation system based on state update, which is used to realize the method of the first aspect of the present application, and comprises a product real-time tracing subsystem, a product adoption recommendation subsystem and a product consumption recommendation subsystem; the product real-time tracing subsystem comprises a data statistics module, a past archive module, a real-time monitoring module, an environment detection module, a remote operation module and a visual operation interface module; the product adoption recommendation subsystem comprises a consumer tracking adopted product state module, a tracing archive module, a consumer browsing and selecting adoption product module, a recommendation model and algorithm module and a visual operation interface module; and the product consumption recommendation subsystem comprises a mall recommendation model and algorithm module, a consumer behavior record and a visual operation interface module.

[0016] Compared with the prior art, the present application has the following advantages:

[0017] 1.The present application combines the three modes of smart farming, consumer adoption and online mall, and breaks the chain of adoption and consumption in the new background of smart agriculture adoption and traceability. Previous technologies often only consider the properties of the commodity itself or simply include part of the product traceability information as an accessory information of the commodity characteristics, but lack of considering adoption and traceability as the main body. Consumers can choose product adoption through the present application, or directly purchase products with complete growth records in the mall. Adoption and traceability will be the main body rather than the accessory of the mall. The present application can help consumers better choose products from the perspective of safety and traceability, and through the time series information of the growth of agricultural products, consumers can grasp the whole process from growth to sales. The present application solves the problem of consumer trust in the growth process of agricultural products, the problem of recommending products for adoption and the problem of increasing the consumption value of agricultural products by introducing consumers into the growth process of agricultural products. The present application is a feasible and relatively comprehensive recommendation system and method for the integration of agricultural product traceability and adoption, and has certain use value and economic value.

[0018] 2.The recommendation method of the present application considers the time series information of the state of the recommended product. General recommendation methods are for the recommendation of static goods, and the present application realizes a recommendation system and method for dynamic goods by collecting the growth records of agricultural products with time series characteristics and introducing a long short-term memory neural network model. This has two advantages, the first is that the recommendation algorithm considers a longer time range, so that the whole process of agricultural products can be considered, and the second is that the recommendation content can be dynamically adjusted according to the state of the agricultural products. BRIEF DESCRIPTION OF DRAWINGS

[0019] Figure 1 is a product recommendation method flowchart based on state update of an embodiment of the present application;

[0020] Figure 2 is a product recommendation stage diagram based on state update of an embodiment of the present application;

[0021] Figure 3 is a product recommendation system structure diagram based on state update of an embodiment of the present application. DETAILED DESCRIPTION

[0022] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0023] In order to better illustrate the advantages, implementation methods and technical routes of the present application, the specific implementation conditions and modes of the present application will be described below by way of examples. The following examples can cover general cases but cannot represent all cases, and cases not covered in the examples and not substantially modified in the method structure of the present application still fall within the scope of the present application.

[0024] The present application aims to provide a product recommendation method and system based on state updating, which can train a deep learning model using environmental monitoring, real-time monitoring and other index information generated during the adoption process, use the real-time updated adoption product state and comprehensively consider the consumer time preference vector to change the push ranking, and recommend the most suitable and most interesting adopted agricultural products or agricultural product commodities to the consumer. The present application can not only provide personalized and accurate services for consumers when they adopt agricultural products, but also provide consumers with safer and higher quality agricultural products by comprehensively considering the agricultural product traceability when they select and purchase agricultural products.

[0025] In the embodiment of the present application, the present embodiment is divided into three stages, the first stage is real-time traceability, the second stage is pre-sale product recommendation, i.e. adoption recommendation, and the third stage is formal product recommendation, i.e. commodity recommendation; in the first stage, environmental information of the product growth or production process is collected by using environmental monitoring equipment, real-time monitoring information of the product is collected by using monitoring equipment, and other index information of the product is collected by using other sensors. At the same time, past archives are sorted out, displayed through a visual interface, and transmitted through remote operation to transmit the above information, and prepared for the second stage; in the early stage of the second stage, the information of the first stage is sorted out to form product state time series data and consumer preference state time series data, and these data can recommend suitable adoption products to consumers through appropriate model algorithms; in the third stage, since some products have been adopted for pre-sale, and some products have not been adopted for pre-sale, therefore, in this stage, the states of these products and the states of consumers are combined to recommend suitable formal commodities to consumers.

[0026] Figure 1 The product recommendation method based on state updating of the embodiment of the present application is shown in the flowchart as shown in Figure 1 The method comprises:

[0027] 101, obtaining the historical state and the current state of the product growth process to form product state time series data;

[0028] In the embodiment of the present application, the product state time series data can include information containing product attributes and commodity attributes, such as the weight and health state of crops, and the price of the product, etc. By processing the product time series state data, the attribute information of the product can be accurately extracted for subsequent recommendation.

[0029] 102. Obtain historical preference state data and current preference state data of the consumer consumption process to form consumer preference state time series data;

[0030] In the embodiment, steps 101 and 102 are both for obtaining data required by the recommendation process, step 101 is mainly to obtain the current state of the product growth process through environmental monitoring, real-time monitoring and the like, and to obtain historical state information stored in a server / memory and the like through calling; and step 102 is mainly to obtain historical preference state data and current preference state data stored in a server / memory and the like through calling.

[0031] Among them, the real-time product feature vector comprehensively depicts various indexes and information of the agricultural product in the breeding process and the time-series product attributes in the form of a vector. The browsing behavior, adoption record and purchase record of the consumer will be used to generate a real-time consumer preference feature vector to depict the characteristics and preferences of the consumer. The formal product feature vector depicts the information of the product in terms of price, category and the like to quantitatively depict the product features from multiple angles. The real-time product feature vector functions separately with the real-time consumer preference feature vector in the adoption recommendation algorithm, and in the mall, it will function together with the commodity attribute to generate a commodity feature vector to function together with the consumer feature vector.

[0032] 103. Input the product state time series data and the consumer preference state time series data into the recommendation model based on the double-state dynamic neural network to output a product feature vector and a consumer preference feature vector;

[0033] In the embodiment of the application, a double-state dynamic long short-term memory neural network model is adopted, which is trained by using the state data generated in the growth process of the agricultural product and the dynamic consumer preference vector, and outputs a consumer preference product list and a new state product state vector and a consumer preference vector.

[0034] For the product state vector, the method of the application can consider the past time series state information of the product with different weights, so that the current product vector state information can more deeply reflect the overall situation of the product; for the consumer preference vector, the model will consider the past consumer preference vector, and based on this method, the hidden preference behavior can be dynamically captured. On this basis, the model will comprehensively consider the consumer preference vector and the product state vector to generate a product ranking list of interest to the consumer.

[0035] The double-state time series recommendation method of the application is a further improvement based on the existing model structure, and the structure, algorithm formula and training method of the overall model have certain originality. The specific operation and algorithm are as follows:

[0036] c t = σ(wu [a t-1 ,x t ]+b u )*tanh(w c [a t-1 ,x t ]+b c )+σ(w f [a t-1 ,x t ]+b f )*c t-1 (1)

[0037] q vt =σ(w i [v t-1 ,xv t ]+b i )*tanh(c t ) (2)

[0038] a t =σ(w o [a t-1 ,x t ]+b o )*tanh(c t ) (3)

[0039] v t =σ(w s [v t-1 ,xv t ]+b s )*tanh(c t )+q vt (4)

[0040] wherein (1), (2) are intermediate information vector update formula, (3), (4) are product state vector and consumer preference vector update formula respectively. In formula (1), (2), subscript t represents state, a t represents real-time product feature vector at t moment, a t-1 reasonably represents real-time product feature vector at t-1 moment; c t represents intermediate time sequence feature vector at t moment, that is, unit state and product memory state information at t moment, is intermediate variable containing product memory information; q vt represents intermediate variable containing consumer v preference and product memory state information at t moment; xv t represents consumer preference state data, x trepresents the product state data. In formula (3), (4), w, b represent model parameters, subscript f represents the model forget gate part, c represents the candidate memory cell part, u represents the model input gate part, i represents the model fusion gate, o represents the model output gate part, and s represents the consumer gate part, and specific values need to be generated by iterative training. The generation of the product state vector depends on the previous state vector and c t the intermediate variable, and the generation of the consumer preference vector needs to depend on the previous preference vector, c t the intermediate variable, and q vt the intermediate variable containing preference information.

[0041] Compared with the traditional recommendation algorithm, the present application has the characteristics of real-time updating of the adoption product state and dynamically considering the consumer preference, so as to realize real-time pushing according to the product and consumer double states, and can fill the blank of the recommendation system in the field of agricultural product adoption. Compared with the traditional algorithm considering only one time variable, the present application proposes an intermediate variable q vt containing consumer preference information and product state information, and proposes a model fusion gate for calculating q vt This improvement can fuse the characteristics of the product and the consumer, so as to better improve the accuracy of the recommendation. In addition, the traditional algorithm has only one output vector, and the present application simultaneously uses the model fusion gate and the consumer gate to output the consumer preference vector v t and the product state vector a t , so as to better improve the recommendation accuracy and adapt to the product and consumer double time sequence scene.

[0042] 104. Output the real-time product feature vector and the real-time consumer preference feature vector according to step 103, and obtain a pre-sale product recommendation list by using the collaborative filtering method;

[0043] In some embodiments of the present application, it is considered that the product recommended to the consumer should be both diverse and relevant, that is, while maintaining the accuracy of the recommendation result, excellent performance in diversity is also presented. The weight related to the adoption interest degree of the user is increased, so that the diversity weight of the user with a higher adoption interest degree is greater, and the improvement of diversity is more concerned; at the same time, the diversity weight of the user with a lower adoption interest degree is reduced, and the accuracy is preferentially ensured. In this way, the parameter adaptive adjustment for users with different adoption interest degrees is realized, so as to improve the overall recommendation quality. Therefore, the corresponding pre-sale product list can also be determined by the following way:

[0044]

[0045] wherein, Rank (u, i) represents the score of the item of interest of the consumer u, f (u) represents the adoption interest degree of the consumer u, r vi, wherein sim(u,v) represents the similarity between consumer u and consumer v, S represents the consumer set, S(u,k)∩N(i) represents the intersection of the k consumers closest to consumer u and the consumers interested in the i product.

[0046] The weight parameter λ of the traditional MMR algorithm is optimized as the adoption interest degree f(u) of the consumer in the embodiment, and the higher the adoption interest degree is, the higher the diversity of the product selected by the consumer is, and the accuracy of the recommendation result is improved.

[0047] In some embodiments of the present application, the general collaborative filtering method also uses the Pearson correlation measurement method to measure the correlation between users, and the Pearson correlation coefficient is a simplest method that can help to understand the relationship between characteristics and response variables, which measures the linear correlation between variables and is only sensitive to linear relationship, and the result is in the range of [-1, 1], -1 represents complete negative correlation (when this variable decreases, the other variable increases), +1 represents complete positive correlation, and 0 represents no linear correlation; however, when the Pearson correlation coefficient is 0, it cannot be determined that the two variables are independent (they may be nonlinearly correlated); in order to solve this problem, the present application uses the number n of consumers in the intersection of the k consumers closest to consumer u and the consumers interested in the i product, combines the Pearson correlation coefficient between the characteristic user and the user and the Pearson correlation coefficient between the user and the product to calculate the interest ranking, which can be expressed as:

[0048]

[0049] , wherein n represents the number of consumers, i.e., the number of consumers in the intersection of the k consumers closest to consumer u and the consumers interested in the i product; , wherein sim(u,v) represents the similarity between consumer u and consumer v, S represents the consumer set, S(u,k)∩N(i) represents the intersection of the k consumers closest to consumer u and the consumers interested in the i product. vi , wherein sim(u,v) represents the similarity between consumer u and consumer v, S represents the consumer set, S(u,k)∩N(i) represents the intersection of the k consumers closest to consumer u and the consumers interested in the i product.

[0050] In the preferred embodiment of the present application, the collaborative filtering recommendation results obtained by using the improved MMR algorithm and the improved Pearson measurement algorithm can also be weighted and screened, i.e., the recommendation results of formula (5) and formula (6) are weighted to obtain more reasonable recommendation results.

[0051] Based on the above embodiments, the present application can recommend a suitable pre-sale product recommendation list to the consumer.

[0052] 105、According to the pre-sale product recommendation list, a corresponding adoption product is selected.

[0053] In the embodiment, after the user is recommended the pre-sale product recommendation list, the user will complete the adoption based on the pre-sale product recommendation list, and after the adoption is completed, the embodiment will also collect the adoption-related information, i.e., the adoption user information and the adoption product information, to prepare for subsequent formal product recommendation.

[0054] 106. Input the product state time series data and the consumer preference state time series data of the product which is not adopted after the product growth process is completed into the recommendation model based on the double-state dynamic neural network, and output the real-time product feature vector and the real-time consumer preference feature vector;

[0055] In the embodiment, due to the product adoption result, the product which is not adopted will be endowed with some commodity attributes, which can be the conventional attribute information such as the price and the commodity category. At the same time, since the consumer completes the adoption of the product in the adoption stage, the consumer preference feature vector can change, and therefore the embodiment also updates the consumer preference feature vector in this process, so that the real-time product feature vector and the real-time consumer preference feature vector finally output are associated with the adoption stage and the formal recommendation stage.

[0056] 107. According to the real-time product feature vector and the real-time consumer preference feature vector output in step 106, a formal product recommendation list is obtained by using the collaborative filtering method.

[0057] In the embodiment, the product feature vector is generated by using the final time series product vector and the commodity attribute. In this step, the consumer preference product vector can be updated, the commodity attribute information can be changed and cause the commodity feature vector to change, and thus the commodity recommendation result list is changed.

[0058] In the embodiment, similar to step 103, the product feature vector and the consumer preference feature vector are input into the model in the embodiment, and different from step 103, the final formal product feature vector after the adoption completion stage and the updated consumer preference feature vector after the adoption completion stage are input in the embodiment, at this time, the real-time product feature vector and the real-time consumer preference feature vector are also output.

[0059] In the embodiment of the present application, the present application also considers that the consumer will be affected by the anchoring effect, that is, when the heat of a certain product rapidly rises, a series of similar products or derivative products will appear on the market, and when the user contacts these similar products or derivative products, the user is often dominated and influenced by the first contact product, and some adoption experience or purchase experience of the original product will guide the user to quickly make a suitable judgment on the new similar product or derivative product, therefore, the adoption interest degree will also be focused on in this stage, which will greatly affect the user's shopping desire whether to purchase the product, thereby improving the overall recommendation quality.

[0060] It can be understood that the dual-state timing recommendation method of the present application can be used in the product adoption and shopping mall consumption two stages compared with the traditional recommendation algorithm, as shown in the following table. Figure 2

[0061] For the product adoption stage, the product to be adopted is recommended, and at this time the recommendation algorithm considers the time-series product state vector and the consumer preference vector.

[0062] For the shopping mall consumption stage, the product that has completed all stages of product breeding but has not been adopted is recommended, at this time the model will use the final time-series product state vector and the product attribute to jointly generate the product feature vector and the time-series consumer preference vector. At this time, the product vector in the recommendation system will no longer be updated with time, and the consumer preference vector continues to be updated with time.

[0063] For the special case of the first use of the consumer, the present recommendation algorithm will use the cold start method. Because the consumer uses the adoption system or the online shopping mall for the first time, the algorithm model cannot obtain the consumer preference according to the consumer's historical records, so it cannot produce adoption recommendation results or shopping mall recommendation results. In view of this situation, the algorithm model will comprehensively statistics the most adopted agricultural products or goods or the most collected agricultural products or goods in recent times and form a recommendation list.

[0064] Figure 3 The product recommendation system structure based on state update of the embodiment of the present application is shown in the following figure. Figure 3 The system includes a product real-time traceability subsystem, a product adoption recommendation subsystem, and a product consumption recommendation subsystem; the product real-time traceability subsystem includes a data statistics module, a past archive module, a real-time monitoring module, an environment detection module, a remote operation module, and a visual operation interface module; the product adoption recommendation subsystem includes a consumer tracking adopted product state module, a traceability archive module, a consumer browsing and selecting adopted product module, a recommendation model and algorithm module, and a visual operation interface module; the product consumption recommendation subsystem includes a shopping mall recommendation model and algorithm module, a consumer behavior record, and a visual operation interface module.

[0065] ​The environment monitoring module is used to monitor the environment of the agricultural products, the real-time monitoring module is used to observe the growth status of the agricultural products in real time, the data statistics module is used to analyze various indexes of the agricultural products, the past archive module is used to record the whole growth process of the agricultural products, the remote control module is used to remotely manage the equipment, and the visual operation interface module is used by the breeder. The agricultural product adoption sub-system includes a product state tracking module used to view the real-time monitoring of the adopted products in real time, a growth environment and other indexes, a traceability archive module used for consumers to view the historical records of the adopted agricultural products, a consumer browsing and selecting adopted product module used to select agricultural products for adoption, a recommendation model and algorithm module used to provide a recommended adoption list for each consumer, and a visual operation interface module used for the consumer to operate. The online mall sub-system includes a mall recommendation model and algorithm module used to provide a recommended product list for each consumer, a consumer behavior record used to record the consumer preferences, and a visual operation interface module used for the consumer to operate.

[0066] The breeder can generate a traceability archive of the bred agricultural products by collecting parameters through the hardware facilities, analyzing the parameters, and filling in indexes such as the variety and weight in the breeding traceability visual operation interface. For each agricultural product bred by the breeder, the recommendation method considering the time sequence information generates an adoption recommendation list according to the past traceability archive of the agricultural product, the current product status, and a neural network model, and updates the list in real time after the product status changes. The consumer can select an intended adopted product from the recommendation list in the adoption operation interface, and can view the current status and past traceability archive of the product. The consumer can obtain real-time push of the status of the adopted product after adopting the product. The adopted product of the consumer will be arranged for logistics distribution and a product corresponding traceability two-dimensional code.

[0067] A person of ordinary skill in the art can understand that all or part of the steps in the various methods of the above embodiments can be completed by programs instructing related hardware, and the programs can be stored in a computer readable storage medium, which can include ROM, RAM, magnetic disks or optical disks, etc.

[0068] Although the embodiments of the present application have been shown and described, it should be understood by those skilled in the art that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the present application, and the scope of the present application is defined by the appended claims and their equivalents.

Claims

1. A product recommendation method based on state updates, characterized in that, The method includes:

101. Obtain the historical and current states of the product's growth process to form product state time-series data; 102. Obtain historical and current preference status data of consumers during the consumption process to form time-series data of consumer preference status; 103. Input product state time-series data and consumer preference state time-series data into a recommendation model based on a dual-state dynamic neural network, and output real-time product feature vectors and real-time consumer preference feature vectors; the dual-state dynamic neural network is based on a dual-state time-series recommendation method; the dual-state time-series recommendation method includes: ； ； ； in, and This represents an intermediate information vector. This represents a consumer preference vector; the subscript t indicates the state. This represents the product time-series feature vector at time t-1; This represents the intermediate time series feature vector at time t; This indicates that time t includes consumers. Intermediate variables for preference and product memory status information; Data representing consumer preference states, This indicates product status data; In the formula, , The subscripts f and c represent the model forget gate, c represent the candidate memory unit, u represent the model input gate, i represent the model fusion gate, o represent the model output gate, and s represent the consumer gate.

104. Based on the real-time product feature vector and real-time consumer preference feature vector output in step 103, obtain a pre-sale product recommendation list using collaborative filtering.

105. Select the appropriate adoption product from the pre-sale product recommendation list; 106. Input the time series data of the status of products that have not been adopted after the product growth process is completed and the time series data of consumer preference status into the recommendation model based on the dual-state dynamic neural network, and output the real-time product feature vector and the real-time consumer preference feature vector.

107. Based on the real-time product feature vector and real-time consumer preference feature vector output in step 106, obtain the formal product recommendation list using collaborative filtering.

2. The product recommendation method based on state update according to claim 1, characterized in that, The product time-series feature vector is calculated using the following methods: ； in, This represents the real-time product feature vector at time t. This represents the activation function. This represents the weight parameters corresponding to the output gate. This represents the bias parameter corresponding to the output gate; This represents the activation function.

3. The product recommendation method based on state update according to claim 2, characterized in that, The consumer preference vector is calculated in the following ways: ； in, This represents the real-time consumer time-series preference feature vector at time t. This represents an intermediate variable at time t that contains time-series data of consumer preference states and product states. This represents the consumer's preference status data at time t. This represents the weight parameters corresponding to the consumer gate. This represents the bias parameter corresponding to the consumer gate.

4. The product recommendation method based on state update according to claim 3, characterized in that, The intermediate time series feature vector at time t is calculated in the following ways: ； in, , The weight parameters represent the candidate memory units and the forget gate. , Indicates the bias parameters corresponding to candidate memory units and forget gates; This represents the intermediate time series feature vector at time t-1.

5. The product recommendation method based on state update according to claim 3, characterized in that, The time t includes the consumer Intermediate variables for preference and product memory status information are calculated in the following ways: ； in, This represents the weight parameters corresponding to the fusion gate. This represents the bias parameter corresponding to the fusion gate.

6. The product recommendation method based on state update according to claim 1, characterized in that, The collaborative filtering method employs the following calculation methods: ； in, This represents the score of item i that consumer u is interested in. Consumers Adoption interest This represents the preference coefficient of consumer v for item i. Consumers With consumers Similarity Represents a set of consumers. This represents the intersection of the k closest consumers to consumer u and the consumers who are interested in item i.

7. The product recommendation method based on state update according to claim 1, characterized in that, The collaborative filtering method employs the following calculation methods: ； in, This represents the score of item i that consumer u is interested in. This represents the number of consumers, which is the number of consumers in the intersection of the k closest consumers of consumer u and the consumers who are interested in item i. This represents the average similarity between consumer u and other consumers v; This represents the preference coefficient of consumer v for item i.

8. A product recommendation system based on state update, used to implement the product recommendation method based on state update as described in any one of claims 1 to 7, characterized in that, The system includes a real-time product traceability subsystem, a product adoption recommendation subsystem, and a product consumption recommendation subsystem. The real-time product traceability subsystem includes a data statistics module, a historical record module, a real-time monitoring module, an environmental detection module, a remote operation module, and a visual user interface module. The product adoption recommendation subsystem includes a consumer tracking module for adopted products, a traceability record module, a consumer browsing and selection module for adopted products, a recommendation model and algorithm module, and a visual user interface module. The product consumption recommendation subsystem includes an online store recommendation model and algorithm module, consumer behavior records, and a visual user interface module.

Images