Ice and snow tourism product recommendation method and system based on neural collaborative filtering and linear confidence upper bound

Abstract

The invention discloses an ice and snow tourism product recommendation method and system based on neural collaborative filtering and a linear confidence upper bound, relates to the technical field of e-commerce product recommendation, and aims to solve the problem of exposure deviation in an existing ice and snow tourism product recommendation method. According to the method, the exposure deviation problem existing in ice and snow tourism product recommendation is fully considered, excessive recommendation of hot products can cause a long tail phenomenon of recommendation, flow is concentrated on high-exposure products, and the selection range of a user is limited. Meanwhile, low-exposure products are difficult to enter a recommendation list of the user due to the influence of exposure deviation, so that unfair recommendation is formed. The real preference of the user is captured in combination with multi-feature information of the user, and the influence of exposure deviation on the recommendation result is reduced. The multi-attribute information of the user is extracted by designing the neural collaborative network, and the reward value features are generated by adopting the linear confidence upper bound algorithm, so that the precision and personalization of recommendation are kept while the exploration capability of low-exposure products is improved. According to the method, the user multi-feature information, the product attribute data and the reward value features are fused into the recommendation model, so that the exposure deviation problem in a traditional recommendation model is optimized, and the fairness and diversity of recommendation are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of e-commerce product recommendation, and in particular to a method for recommending ice and snow tourism products based on neural collaborative filtering and a linear confidence upper bound. Technical Background

[0002] In today's digital age, the digital empowerment and cultural tourism integration development model represented by tourism e-commerce is an important way to realize the modernization of Chinese tourism, and the tourism e-commerce industry has ushered in unprecedented development opportunities. With the growing demand for personalized tourism experience, the role of ice and snow tourism product recommendation methods has become particularly important. Especially driven by the popularization of winter sports and the dissemination of ice and snow culture, ice and snow tourism has shown great economic potential. However, the existing ice and snow tourism product recommendation methods still have certain shortcomings in the field of ice and snow tourism. At this stage, the recommendation methods tend to recommend popular ice and snow tourism products, while ignoring the personalized needs of users and potential niche markets, making it difficult for users to obtain more personalized and targeted ice and snow tourism product choices, thereby limiting the user's exposure to niche ice and snow tourism products. In addition, the lack of recommendation of long-tail products has exacerbated the unfairness of information filtering, which not only affects the diversity of ice and snow tourism products, but also limits the optimization of the platform's sales strategy in the ice and snow tourism market. Therefore, the existing recommendation methods have not yet effectively solved the following problems: 1) It is difficult to effectively integrate user behavior data and product attribute data, resulting in a lack of personalization in the recommendation results. 2) The accuracy of preference assessment for new users or new products is not high, which makes the recommendation system perform poorly when facing emerging markets.

[0003] The prior art with document number CN116738066B discloses a rural tourism service recommendation method, device, electronic device and storage medium, which includes: obtaining tourism resources and tourist data; constructing tourism resource portraits; performing sentiment analysis on feedback data to obtain sentiment analysis results; determining a scoring matrix based on feedback data and tourism resource portraits, constructing tourist portraits based on basic data and tourism resource portraits, and tourist portraits including basic feature labels and feature preference labels; calculating the potential similarity, basic similarity and preference similarity of the first tourist based on the scoring matrix, the basic feature labels and feature preference labels of the tourist portraits, and generating a similarity matrix of the first tourist by weighted fusion of the three similarities; determining the first tourist who is similar to the second tourist based on the similarity matrix, and recommending tourism resources to the second tourist. The prior art realizes the reliability and accuracy of tourism resource-related data; and provides tourists with more accurate rural tourism resource recommendation services. The prior art does not effectively integrate user behavior data and product attribute data.

[0004] The prior art with the document number CN115438871A discloses a method and system for recommending ice and snow attractions by integrating preferences and eliminating popularity deviations. The technical key points include: grouping tourists by gender, age, and occupation respectively, and mapping them into values ​​of [0-1]; calculating the similarity of tourists' multi-feature preferences according to the mapped values ​​of tourists' gender, age, and occupation; calculating the tourists' preference values ​​according to the similarity of tourists' multi-feature preferences and tourists' historical rating data; constructing a tourist-preference value matrix according to the standardized tourist preference values; training a matrix decomposition model based on tourists' multi-feature preferences according to the tourist-preference value matrix and tourists' historical rating data to obtain the trained tourist-preference value matrix; predicting and recommending new users according to the trained tourist-preference value matrix. It effectively alleviates the unfavorable situation that low-popularity ice and snow attractions are difficult to be recommended to tourists. However, there is no in-depth consideration of the preferences of new users or new products. Summary of the invention

[0005] In view of the above problems, the present invention proposes a snow and ice tourism product recommendation method based on neural collaborative filtering and linear confidence bound to solve the exposure bias problem in the existing snow and ice tourism product recommendation method.

[0006] A method for recommending ice and snow tourism products based on neural collaborative filtering and linear confidence bounds includes the following steps:

[0007] Step 1: Collect user characteristic information and user historical rating data. User characteristic information includes gender, age and occupation; user historical rating data includes ratings of tourist attractions and browsing, clicking and purchasing records of related e-commerce products;

[0008] Step 2: Collect characteristic information of ice and snow tourism products;

[0009] Furthermore, it includes the ID and type of ice and snow tourism products, ice and snow tourism product packages (whether the product provides packages for multiple ice and snow activities), winter theme activities (whether the product includes winter special activities such as skiing, snowmobiling, ice sculpture exhibitions), traffic convenience (traffic conditions around the product), guided tour services (whether the product provides detailed tour guide explanations or online guided tours), scenic area guidance signs (whether clear guidance signs are set up in the scenic area), special catering (whether local special winter catering is provided), barrier-free facilities (whether the product is equipped with barrier-free facilities), etc.;

[0010] Step 3: Construct feature vectors of users and ice and snow tourism products. Using the user attribute information and ice and snow tourism product attribute information collected in steps 1 and 2, encode each feature into a low-dimensional dense vector through the embedding layer to form user feature vectors and ice and snow tourism product feature vectors.

[0011] Furthermore, for user characteristics, information such as gender, age, and occupation are encoded into feature vectors; for ice and snow tourism product characteristics, attributes such as product ID, product type, activity type (such as skiing, ice sculpture exhibition, etc.), transportation convenience, and guide services are integrated into a single high-dimensional vector representation;

[0012] Step 4: Calculate similarity by combining user behavior characteristics and content characteristics to obtain user preferences;

[0013] Step 5: Introduce the Linear Confidence Upper Bound (LinUCB) algorithm to calculate the reward features, and through dynamic balance between exploration and utilization, discover products that users may be interested in but have low exposure rates;

[0014] Furthermore, the LinUCB algorithm calculates the potential reward value of each ice and snow tourism product through historical user behavior and product attributes. In each recommendation, the system prioritizes products for recommendation based on the upper bound of the confidence interval of the reward value, thereby increasing the recommendation frequency of low-exposure ice and snow tourism products.

[0015] The expected rewards for each user for each ice and snow tourism product are as follows:

[0016]

[0017] b represents the set of ice and snow tourism products to be selected; represents the feature vector corresponding to the ice and snow tourism product b in the hth experiment, and the preference vector is unknown, but its estimated value can be calculated by ridge regression in each round of interaction. Each ice and snow tourism product maintains a

[0018] Step 6: Embed the reward value data into the neural collaborative filtering network model to generate reward value embedding vectors for users and products, and combine them with their respective nearest neighbor feature vectors to form the user and product representations required for matrix decomposition and multi-layer perceptron.

[0019] Step 7: The user and ice and snow tourism product representations obtained in step 6 are used to learn the interaction function between the user and the ice and snow tourism product, and obtain the output vector of the multilayer perceptron

[0020]

[0021] Among them, ReLU is the activation function, Q f is the weight matrix of the f-th layer, is the output of layer f-1, e f is the bias vector.

[0022] Step 8: Concatenate the output vectors of the matrix decomposition and the multi-layer perceptron and input them into a fully connected layer to obtain the predicted score between user p, ice and snow tourism product s and reward value u:

[0023]

[0024] Among them, h is the weight vector of the output layer, e represents the bias term of the output layer, and σ(·) is the Sigmoid function.

[0025] Step 9: The output layer completes the prediction task and optimizes the output of the user's preference probability for ice and snow tourism products according to the binary cross entropy loss function. The prediction is converted into a binary classification problem, and the prediction probability range is set to [0,1];

[0026] Step 10: Sort the final preference probabilities obtained in step 9 from high to low, and select the top t ice and snow tourism products for recommendation, where t represents the number of recommended ice and snow tourism products.

[0027] Step 11: Use standard data sets to evaluate the performance of the method of the present invention, and conduct ablation experiments, comparative experiments, and hyperparameter experiments respectively.

[0028] The beneficial technical effects of the present invention are:

[0029] The present invention combines neural collaborative filtering and linear confidence bound algorithm and introduces reward value feature to effectively solve the exposure bias problem in the current ice and snow tourism product recommendation method. By obtaining the user's gender, age, occupation and other feature information and historical rating data, and mapping this information into a value between [0 and 1], a user-multi-feature matrix is ​​constructed to better capture the user's true preferences, thereby reducing the impact of exposure bias on the recommendation results.

[0030] The present invention fully considers the exposure bias problem in the recommendation of ice and snow tourism products. Over-recommendation of popular products will lead to a long-tail phenomenon in the recommendation, which will concentrate the traffic on high-exposure products and limit the user's selection range. At the same time, due to the influence of exposure bias, it is difficult for low-exposure products to enter the user's recommendation list, thus forming an unfair recommendation. The present invention combines user multi-feature information to capture the user's real preferences and reduce the impact of exposure bias on the recommendation results. By designing a neural collaborative network to extract user multi-attribute information and using a linear confidence upper bound algorithm to generate reward value features, the accuracy and personalization of recommendations are maintained while improving the exploration ability of low-exposure products. The present invention integrates user multi-feature information with product attribute data and reward value features into the recommendation model to optimize the exposure bias problem in the traditional recommendation model and improve the fairness and diversity of recommendations. Finally, the present invention not only improves the overall performance of the recommendation system, but also enhances the user experience, providing a more scientific and accurate recommendation solution for the tourism e-commerce platform.

[0031] The present invention designs a neural collaborative network to extract user multi-attribute information, and combines the attribute information of ice and snow tourism products, including product ID, type, package, winter theme activities, traffic convenience, guide service, scenic spot guide signs, special catering, barrier-free facilities, etc., and converts these attribute information into binary sparse vectors through one-hot encoding and inputs them into the embedding layer to generate embeddings. These feature vectors are respectively input into the multi-layer perceptron module and the matrix decomposition module, and then the element product is output through the fully connected layer and the softmax output layer. Finally, the output layer completes the prediction task, outputs the user's preference probability for ice and snow tourism products, and uses the binary cross entropy loss function for optimization, which solves the problem that the recommendation system performs poorly when facing new users or new products, optimizes the deep mining of user behavior and product characteristics, and improves the recommendation ability for low-exposure products, effectively alleviates the unfairness of information filtering of long-tail ice and snow tourism products, and improves the fairness and diversity of recommendations, so that the recommendation system can better meet the personalized needs of users and promote the development of the tourism e-commerce industry.

[0032] The present invention effectively integrates user behavior data and product attribute data to personalize the recommendation results. The present invention realizes accurate evaluation of the preferences of new users or new products, making the recommendation system highly adaptable in the face of emerging markets. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 A schematic flow chart of the present invention is shown;

[0034] Figure 2 This is a comparison chart showing that the present invention exhibits the best fairness and the best algorithm performance in the recommendation stage (it can be seen that the present invention has the smallest Gini coefficient compared with other methods);

[0035] Figure 3 This is another comparison chart showing the best fairness and algorithm performance of the present invention in the recommendation stage (it can be seen that the present invention has the smallest Gini coefficient compared with other methods);

[0036] Figure 4 This is an experimental result diagram that compares the performance of the model on the dataset with different learning rates. DETAILED DESCRIPTION

[0037] Combined with the chart, the specific steps of the ice and snow tourism product recommendation method based on neural collaborative filtering and linear confidence upper bound of the present invention include:

[0038] The first step is to collect users' historical ratings of ice and snow tourism products and their gender, age and occupation information provided when registering. Based on this information, users are preliminarily classified, and users with the same gender, age or occupation are divided into the same group. The gender, age and occupation characteristics of users are mapped to the [0-1] interval.

[0039] The second step is to collect characteristic information of ice and snow tourism products. The ice and snow tourism product ID is the unique identifier of each product, which helps the system to accurately track and analyze the exposure and user interaction of ice and snow tourism products. Type information is an important factor for users to choose ice and snow tourism products, which can improve the relevance of recommendations and user satisfaction.

[0040] Furthermore, the characteristic information of the product also includes the ID and type of the ice and snow tourism product, ice and snow tourism product packages (whether the product provides packages for multiple ice and snow activities to meet the comprehensive experience needs of users), winter theme activities (whether the product includes winter special activities such as skiing, snowmobiling, ice sculpture exhibitions, etc. to enrich the user's ice and snow tourism experience), traffic convenience (traffic conditions around the product, such as whether direct or convenient shuttle transportation is provided to facilitate users to go to ice and snow tourist attractions), guided services (whether the product provides detailed tour guide explanations or online guided tours to help users deeply understand the culture and characteristics of ice and snow tourist attractions), scenic spot guidance signs (whether clear guidance signs are set up in the scenic area to help users smoothly tour in the ice and snow environment), special catering (whether local special winter catering is provided to meet users' needs for local cuisine), barrier-free facilities (whether the product is equipped with barrier-free facilities to provide barrier-free travel experience for users with limited mobility), etc.;

[0041] The third step is to use the user attribute information and ice and snow tourism product attribute information collected in step 1 and step 2 to construct the feature vectors of users and ice and snow tourism products. Each feature is encoded into a low-dimensional dense vector through the embedding layer to form the user feature vector and the ice and snow tourism product feature vector;

[0042] Furthermore, for user characteristics, information such as gender, age and occupation is encoded into feature vectors. The occupation information of users is mapped through a predefined occupation dictionary and converted into a one-hot encoding vector, and the user's age is directly input into the model as a continuous value feature; for ice and snow tourism product characteristics, attributes such as product ID, type, activity type (such as skiing, ice sculpture exhibition, etc.), transportation convenience, and guide services are integrated into a single high-dimensional vector representation;

[0043] Among them, after obtaining the feature vectors of users and ice and snow tourism products in the third step, considering the need to optimize model performance and maintain reasonable computing efficiency, we choose to reduce the dimension of this vector through a fully connected layer and process it through a ReLU activation function to obtain the final feature representation of the product.

[0044] The fourth step is to calculate the similarity by combining the user's behavior characteristics and content characteristics to obtain user preferences;

[0045] The fifth step is to introduce the linear confidence upper bound (LinUCB) algorithm to calculate the reward features, and through dynamic balance between exploration and utilization, to discover products that users may be interested in but have low exposure;

[0046] Furthermore, the LinUCB algorithm calculates the potential reward value of each ice and snow tourism product through historical user behavior and product attributes. In each recommendation, the system prioritizes products based on the upper bound of the confidence interval of the reward value, thereby increasing the recommendation frequency of low-exposure ice and snow tourism products. The expected reward for each user for each ice and snow tourism product is as follows:

[0047]

[0048] b represents the set of ice and snow tourism products to be selected; represents the feature vector corresponding to the ice and snow tourism product b in the hth experiment, and the preference vector is unknown, but its estimated value can be calculated by ridge regression in each round of interaction. Each ice and snow tourism product maintains a The system will eventually recommend the ice and snow tourism product with the highest upper bound of the confidence interval:

[0049]

[0050] in B b is a c*c matrix composed of historical information, c is the dimension of the context feature vector, C b represents the input training data, R b represents the estimated parameter, and α is a hyperparameter that balances exploration-exploitation.

[0051] The historical information mainly includes the interaction records between users and items (such as ratings), context features (age and occupation), and a feature set consisting of parameter estimation values ​​during the model training process.

[0052] Furthermore, the ridge regression method is used to calculate the optimal reward value, and the optimization objective is as follows:

[0053]

[0054] Among them, Rewards is the result of h trials, and the reward value is defined accordingly Finally, the system will record the item with the maximum reward value, and Rewards = 1 and Rewards = 0 represent the user's acceptance and rejection of the recommendation respectively. h ] T is the matrix of eigenvectors observed in h trials, where each row corresponds to an eigenvector. That is, L2 regularization, where I is a diagonal matrix, can effectively avoid overfitting.

[0055] right Taking the derivative, we get:

[0056]

[0057] If the bth ice and snow tourism product has been recommended by f b times, the probability of the product receiving a reward can be calculated as:

[0058]

[0059] For any D>0, there exists:

[0060]

[0061] Among them, P is the true probability of the product, and ω is a hyperparameter. Therefore, the selection mechanism of the LinUCB algorithm satisfies:

[0062]

[0063] in,

[0064] The sixth step is to embed the reward value data into the neural collaborative filtering network model to generate the reward value embedding vectors of users and products, and combine them with their respective nearest neighbor feature vectors to form the user and product representations required by matrix decomposition and multi-layer perceptron;

[0065] The seventh step is to obtain the user and ice and snow tourism product representations corresponding to the matrix decomposition and multi-layer perceptron. The bottom input layer consists of seven features, namely user ID, user gender, user age, user occupation, ice and snow tourism product ID, ice and snow tourism product type and reward value features. The interaction function between users and ice and snow tourism products is learned to obtain the output vector of the multi-layer perceptron.

[0066]

[0067] Among them, ReLU is the activation function, Q f is the weight matrix of the f-th layer, is the output of layer f-1, e f is the bias vector.

[0068] The eighth step is to concatenate the output vectors of the matrix decomposition and the multi-layer perceptron, input them into a fully connected layer, and obtain the predicted score between the user p and the ice and snow tourism product s and the reward value u:

[0069]

[0070] Among them, h is the weight vector of the output layer, e represents the bias term of the output layer, and σ(·) is the Sigmoid function.

[0071] The ninth step is to complete the prediction task in the output layer, output the user's preference probability for ice and snow tourism products, and use the binary cross entropy loss optimization function. The positive sample indicates that there is interaction between the user and the ice and snow tourism product, and the negative sample indicates that there is no interaction between the user and the ice and snow tourism product. The prediction is converted into a binary classification problem, and the prediction probability range is set to [0,1]:

[0072]

[0073] Among them, y represents the positive sample set, y - Represents the set of negative samples.

[0074] The tenth step is to sort the final preference probabilities obtained in the ninth step from high to low, and select the top t ice and snow tourism products for recommendation, where t represents the number of recommended ice and snow tourism products.

[0075] The eleventh step is to use standard data sets to evaluate the performance of the method of the present invention, and conduct ablation experiments, comparative experiments and hyperparameter experiments respectively.

[0076] Furthermore, the ablation experiment is to verify the effectiveness of each component in the method of the present invention. The neural collaborative filtering component and the reward value feature fusion part in the method are removed respectively, and the ablation experiment is performed on the MovieLens-100K and MovieLens-1M datasets. The results are shown in Table 1;

[0077] Table 1

[0078]

[0079] Furthermore, by comparing the traditional NCF, EBPR, FaiRIR_RL, FaiRIR_Sim and the method of the present invention, their recommendation performance is evaluated using fairness indicators and accuracy indicators. The results show that on different data sets, the exposure performance of the five models varies, but they all help to reduce bias.

[0080] The NCF model has the lowest exposure under all recommended numbers; the EBPR model has a relatively stable exposure and has improved, proving that it has the ability to alleviate exposure bias; the FaiRIR_Sim model has significantly improved exposure when the recommended numbers are 10 and 20, showing an initial advantage; the FaiRIR_RL model has a high overall exposure, reflecting the effectiveness of optimizing exposure balance; the method of the present invention has the highest exposure under all recommended numbers, which is about 60% higher than that of the NCF model, effectively reducing exposure bias.

[0081] Furthermore, the fairness of the model is evaluated based on the Gini coefficient (the lower the Gini coefficient, the higher the fairness). In the MovieLens-100K dataset, the NCF model has the lowest fairness and the highest Gini coefficient; the EBPR, FaiRIR_Sim and FaiRIR_RL models perform better when the number of recommendations is low, especially the FaiRIR_RL model; the method of the present invention shows the best fairness in all recommendation stages. The results are shown in Figure 2 , 3 ;

[0082] Furthermore, in the model training process, the learning rate (lr) is a key hyperparameter that has a decisive influence on the convergence speed of the model and the final recommendation performance. In order to accurately explore the impact of the learning rate on the model performance, the method of the present invention uses a grid search strategy to optimize a set of predefined learning rate values ​​{0.00001, 0.0001, 0.0005, 0.001, 0.005}. Under the condition of keeping other parameters unchanged, the performance of the model on the data set with different learning rates is compared experimentally. The experimental results are as follows: Figure 4 shown.

[0083] Experiments have proven that the neural collaborative filtering model that integrates reward features is superior to the original model in terms of alleviating exposure bias and improving recommendation accuracy. While improving the ability to explore low-exposure products, it maintains the accuracy and diversity of recommendations, covers recommendation options more comprehensively, reduces biased recommendations for high-exposure products, and significantly alleviates the exposure bias problem, thereby achieving fairness in recommendation results.

[0084] It can be seen that the present invention fully considers the exposure bias problem in the recommendation of ice and snow tourism products. Over-recommendation of popular products will lead to a long-tail phenomenon in the recommendation, which will concentrate the traffic on high-exposure products and limit the user's selection range. At the same time, due to the influence of exposure bias, low-exposure products are difficult to enter the user's recommendation list, thus forming unfair recommendations. The present invention combines user multi-feature information to capture the user's true preferences and reduce the impact of exposure bias on recommendation results. By designing a neural collaborative network to extract user multi-attribute information and using a linear confidence upper bound algorithm to generate reward value features, the accuracy and personalization of recommendations are maintained while improving the ability to explore low-exposure products.

[0085] The present invention integrates user multi-feature information with product attribute data and reward value features into the recommendation model to optimize the exposure bias problem in the traditional recommendation model and improve the fairness and diversity of recommendations.

[0086] It has been verified that the method proposed in the present invention solves the technical problem proposed in the present invention. The method of the present invention has been verified through practical application of the technical effect and practicality claimed by the present invention.

[0087] The method of the present invention has been verified through simulation experiments and practical applications, and the technical effects claimed by the present invention have been verified.

[0088] The algorithm (method) proposed in the present invention is the underlying technical core of the present invention, and various products can be derived based on the algorithm.

[0089] Based on the algorithm (method) proposed in the present invention, a system for recommending ice and snow tourism products based on neural collaborative filtering and linear confidence bound is developed using a programming language. The system has program modules corresponding to the steps of the above-mentioned technical solution, and executes the steps of the above-mentioned method for recommending ice and snow tourism products based on neural collaborative filtering and linear confidence bound during operation.

[0090] The computer program of the developed system (software) is stored on a computer-readable storage medium, and the computer program is configured to implement the steps of the above-mentioned method for recommending ice and snow tourism products based on neural collaborative filtering and linear confidence upper bound when called by a processor. That is, the present invention is materialized on a carrier to become a computer program product.

[0091] A device for recommending ice and snow tourism products based on neural collaborative filtering and linear confidence bounds, the device comprising at least one processor and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, the instructions are executed by the at least one processor so that the at least one processor can execute the above-mentioned method for recommending ice and snow tourism products based on neural collaborative filtering and linear confidence bounds, thereby realizing scientific and effective recommendations for ice and snow tourism products on the Internet.

[0092] Various implementations of the systems and techniques described herein can be realized in digital electronic circuit systems, integrated circuit systems, dedicated ASICs (application specific integrated circuits), computer hardware, firmware, software, and / or combinations thereof. These various implementations can include: being implemented in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0093] The computer programs (also referred to as programs, software, software applications, or codes) of the present invention include machine instructions for programmable processors, and these computer programs can be implemented using high-level procedural and / or object-oriented programming languages, and / or assembly / machine languages. As used in the present invention, the terms "machine-readable medium" and "computer-readable medium" refer to any computer program product, device, and / or device (e.g., disk, optical disk, memory, programmable logic device PLD) for providing machine instructions and / or data to a programmable processor, including a machine-readable medium that receives machine instructions as machine-readable signals. The term "machine-readable signal" refers to any signal for providing machine instructions and / or data to a programmable processor.

[0094] It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps recorded in this application can be executed in parallel, sequentially or in different orders, as long as the expected results of the technical solution disclosed in this application can be achieved, they are all within the scope of protection of the present invention.

Claims

1. A method for recommending ice and snow tourism products based on neural collaborative filtering and linear confidence bound, characterized in that: The following steps are involved: Step 1: Collect user characteristic information and user historical rating data: User characteristic information includes gender, age and occupation; user historical rating data includes ratings of tourist attractions and browsing, clicking and purchasing records of related e-commerce products; Step 2: Collect characteristic information of ice and snow tourism products; Step 3: Construct feature vectors of users and ice and snow tourism products. Use the user attribute information and ice and snow tourism product attribute information collected in steps 1 and 2 to encode each feature into a low-dimensional dense vector through the embedding layer to form user feature vectors and ice and snow tourism product feature vectors. Step 4: Calculate similarity by combining user behavior characteristics and content characteristics to obtain user preferences; Step 5: Introduce the Linear Confidence Upper Bound (LinUCB) algorithm to calculate the reward features, and through dynamic balance between exploration and utilization, discover products that users may be interested in but have low exposure rates; Step 6: Embed the reward value data into the neural collaborative filtering network model to generate the reward value embedding vectors of users and products, and combine them with their respective nearest neighbor feature vectors to form the user and product representations required by matrix decomposition and multi-layer perceptron; Step 7: The user and ice and snow tourism product representations obtained in step 6 are used to learn the interaction function between the user and the ice and snow tourism product, and the output vector of the multi-layer perceptron is obtained; Step 8: Concatenate the output vectors of the matrix decomposition and the multi-layer perceptron and input them into a fully connected layer to obtain the predicted scores between the user and the ice and snow tourism products and reward values: Step 9: The output layer completes the prediction task and optimizes the output of the user's preference probability for ice and snow tourism products according to the binary cross entropy loss function. The prediction is converted into a binary classification problem, and the prediction probability range is set to [0,1]; Step 10: Sort the final preference probabilities obtained in step 9 from high to low, and select the top t ice and snow tourism products for recommendation, where t represents the number of recommended ice and snow tourism products.

2. The ice and snow tourism product recommendation method based on neural collaborative filtering and linear confidence bound according to claim 1 is characterized in that: In step 1, users are preliminarily classified, and users with the same gender, age or occupation are divided into the same group. The gender, age and occupation characteristics of users are mapped to the interval [0-1].

3. The ice and snow tourism product recommendation method based on neural collaborative filtering and linear confidence bound according to claim 1 or 2, characterized in that: In step 2, The ice and snow tourism product ID serves as the unique identifier of each product, helping the system to accurately track and analyze the exposure and user interaction of ice and snow tourism products. The type information, as an important factor for users to choose ice and snow tourism products, can improve the relevance of recommendations and user satisfaction. The characteristic information of ice and snow tourism products includes the ID and type of ice and snow tourism products, ice and snow tourism product packages, winter theme activities, transportation convenience, guided tour services, scenic area guide signs, special catering, and barrier-free facilities.

4. The method for recommending ice and snow tourism products based on neural collaborative filtering and linear confidence bound according to claim 3 is characterized in that: In step three, For user characteristics, information such as gender, age and occupation is encoded into feature vectors; for ice and snow tourism product characteristics, attributes such as product ID, type, activity type, transportation convenience, and guide services are integrated into a single high-dimensional vector representation. After obtaining the feature vectors of users and ice and snow tourism products in the third step, considering the need to optimize model performance and maintain reasonable computing efficiency, we choose to reduce the dimension of this vector through a fully connected layer and process it through a ReLU activation function to obtain the final feature representation of the product.

5. The method for recommending ice and snow tourism products based on neural collaborative filtering and linear confidence bound according to claim 4 is characterized in that: In step 5, the LinUCB algorithm calculates the potential reward value of each ice and snow tourism product based on historical user behavior and product attributes. In each recommendation, the system prioritizes products based on the upper bound of the confidence interval of the reward value, thereby increasing the recommendation frequency of low-exposure ice and snow tourism products. The expected rewards for each user for each ice and snow tourism product are as follows: b represents the set of ice and snow tourism products to be selected; represents the feature vector corresponding to the ice and snow tourism product b in the hth experiment, and the preference vector is unknown, but its estimated value can be calculated by ridge regression in each round of interaction. Each ice and snow tourism product maintains a The system will eventually recommend the ice and snow tourism product with the highest upper bound of the confidence interval: in B b is a c*c matrix composed of historical information, c is the dimension of the context feature vector, C b represents the input training data, R b represents the estimated parameter, and α is a hyperparameter that balances exploration-exploitation. The historical information mainly includes the interaction records between users and items (such as ratings), context features (age and occupation), and a feature set consisting of parameter estimation values ​​during the model training process. Furthermore, the ridge regression method is used to calculate the optimal reward value, and the optimization objective is as follows: Among them, Rewards is the result of h trials, and the reward value is defined accordingly Finally, the system will record the item with the maximum reward value, and Rewards = 1 and Rewards = 0 represent the user's acceptance and rejection of the recommendation respectively. h ] T is the matrix of eigenvectors observed in h trials, where each row corresponds to an eigenvector. That is, L2 regularization, where I is a diagonal matrix, can effectively avoid overfitting. right Taking the derivative, we get: If the bth ice and snow tourism product has been recommended by f b times, the probability of the product receiving a reward can be calculated as: For any D>0, there exists: Among them, P is the true probability of the product, and ω is a hyperparameter. Therefore, the selection mechanism of the LinUCB algorithm satisfies: in, 6. The method for recommending ice and snow tourism products based on neural collaborative filtering and linear confidence bound according to claim 5, characterized in that: The output vector in step seven is: Step seven is to obtain the user and ice and snow tourism product representations corresponding to the matrix decomposition and multi-layer perceptron. The bottom input layer consists of seven features, namely user ID, user gender, user age, user occupation, ice and snow tourism product ID, ice and snow tourism product type and reward value features. The interaction function between users and ice and snow tourism products is learned to obtain the output vector of the multi-layer perceptron. Among them, ReLU is the activation function, Q f is the weight matrix of the f-th layer, is the output of layer f-1, e f is the bias vector.

7. The method for recommending ice and snow tourism products based on neural collaborative filtering and linear confidence bound according to claim 6, characterized in that: In step eight, the output vectors of the matrix decomposition and the multi-layer perceptron are concatenated and input into a fully connected layer to obtain the predicted score between the user p and the ice and snow tourism product s and the reward value u: Among them, h is the weight vector of the output layer, e represents the bias term of the output layer, and σ(·) is the Sigmoid function.

8. The method for recommending ice and snow tourism products based on neural collaborative filtering and linear confidence bound according to claim 7, characterized in that: Step nine is that the output layer completes the prediction task and outputs the user's preference probability for ice and snow tourism products. The binary cross entropy loss optimization function is used. The positive sample indicates that there is interaction between the user and the ice and snow tourism product, and the negative sample indicates that there is no interaction between the user and the ice and snow tourism product. The prediction is converted into a binary classification problem, and the prediction probability range is set to [0,1]: Among them, y represents the positive sample set, y - Represents the set of negative samples.

9. A snow tourism product recommendation system based on neural collaborative filtering and linear confidence bound, characterized by: The system has a program module corresponding to the steps of any one of claims 1 to 8 above, and executes the steps in the ice and snow tourism product recommendation method based on neural collaborative filtering and linear confidence upper bound when running.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and the computer program is configured to implement the steps of a method for recommending ice and snow tourism products based on neural collaborative filtering and linear confidence bounds as described in any one of claims 1 to 8 when called by a processor.

Images