Product recommendation method and device, electronic equipment and storage medium

The product recommendation method addresses the limitations of sequential decision-based systems by using the large deviation theory to dynamically optimize the threshold multi-armed bandit problem, improving recommendation accuracy and robustness across diverse scenarios.

CN119090586APending Publication Date: 2024-12-06GUANGDONG LAB OF ARTIFICIAL INTELLIGENCE & DIGITAL ECONOMY (SZ)
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411177345.4
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-26
Publication Date
2024-12-06

AI Technical Summary

Technical Problem

Existing recommendation systems, particularly those based on sequential decision-making, suffer from poor applicability and robustness due to reliance on specific assumptions about reward distribution, difficulty in adapting to new users or products, and suboptimal recommendation results under limited resources.

Method used

A product recommendation method utilizing the large deviation theory to dynamically determine candidate products in each round, optimizing the threshold multi-armed bandit problem to directly minimize the loss function, reducing reliance on predetermined parameters, and adapting to user behaviors and product characteristics.

Benefits of technology

Improves recommendation accuracy and robustness across various scenarios, especially in cold start situations, by directly optimizing the loss function and reducing computational complexity, thus enhancing flexibility and efficiency.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119090586A_ABST
    Figure CN119090586A_ABST
Patent Text Reader

Abstract

The embodiment of the invention discloses a product recommendation method and device, electronic equipment and a storage medium, and the method comprises the steps: in a first recommendation stage, determining candidate products recommended to a user in each recommendation round from all candidate products based on a large deviation theory, recommending the corresponding candidate products, and obtaining corresponding returns; obtaining a return of each recommended round of each candidate product at the end of the first recommendation stage to obtain a sample return of each candidate product, and determining an expected return of each candidate product based on the sample return of each candidate product; and in a second recommendation stage, classifying each candidate product as a recommendable product or a non-recommendable product based on the expected return, and recommending the recommendable product to the user. According to the embodiment of the invention, the applicability and robustness to different scenes can be improved, and a more accurate and reasonable recommendation result can be obtained.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to a product recommendation method, device, electronic device and storage medium. Background Art

[0002] Recommendation systems are widely used in e-commerce, content distribution, advertising and other fields, aiming to recommend products such as goods and services that users are interested in. In the prior art, recommendation systems recommend products to users in the following ways: recommendation methods based on user portraits, recommendation methods based on user social network information, and recommendation methods based on sequential decision making.

[0003] However, the sequential decision-based recommendation method in the prior art has the characteristics of relying on given parameters and being based on the fact that the reward distribution conforms to specific assumptions. Therefore, under limited resources, the sequential decision-based recommendation method in the prior art has poor applicability and robustness to different scenarios, and cannot obtain reasonable recommendation results. Summary of the invention

[0004] Embodiments of the present invention provide a product recommendation method, device, electronic device, and storage medium, which can improve the applicability and robustness to different scenarios and obtain more accurate and reasonable recommendation results.

[0005] In a first aspect, an embodiment of the present invention provides a product recommendation method, comprising:

[0006] In the first recommendation stage, based on the large deviation theory, the candidate products recommended to users in each recommendation round are determined from the candidate products, and the corresponding candidate products are recommended and corresponding rewards are obtained;

[0007] Obtaining the returns of each recommended round of each candidate product at the end of the first recommendation stage to obtain sample returns of each candidate product, and determining the expected returns of each candidate product based on the sample returns of each candidate product; and

[0008] In the second recommendation stage, each candidate product is classified as a recommendable product or a non-recommended product based on the expected return, and the recommendable product is recommended to the user.

[0009] In a second aspect, an embodiment of the present invention provides a product recommendation device, including:

[0010] The first-stage recommendation module is used to determine the candidate products recommended to users in each recommendation round from various candidate products in the first recommendation stage based on the large deviation theory, recommend the corresponding candidate products and obtain corresponding rewards;

[0011] an expected return acquisition module, configured to acquire the returns of each recommended round of each candidate product at the end of the first recommendation stage to obtain sample returns of each candidate product, and determine the expected return of each candidate product based on the sample returns of each candidate product; and

[0012] The second stage recommendation module is used to classify the candidate products into recommendable products or non-recommended products based on the expected return, and recommend the recommendable products to the user.

[0013] In a third aspect, an embodiment of the present invention further provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, a product recommendation method as described in any one of the embodiments of the present invention is implemented.

[0014] In a fourth aspect, an embodiment of the present invention further provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements a product recommendation method as described in any one of the embodiments of the present invention.

[0015] A product recommendation method, device, electronic device and storage medium provided in the embodiments of the present invention avoid relying on predetermined additional parameters by determining the products recommended to the user in each recommendation round from each candidate product based on the large deviation theory in the first recommendation stage; and obtaining the returns of each recommended round of each candidate product at the end of the first recommendation stage to obtain the sample returns of each candidate product, and then determining the expected return of each candidate product based on the sample returns of each candidate product, and further obtaining and recommending recommendable products based on whether the expected return is greater than the expected return threshold, which can improve the applicability to different scenarios, especially the cold start scenario where the user is a new user, and improve the robustness of the recommendation, and obtain more accurate and reasonable recommendation results. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solution of the present invention, the drawings required for use in the embodiments are briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0017] Figure 1 is a flowchart of a product recommendation method provided by an embodiment of the present invention;

[0018] Figure 2 is another flowchart of the product recommendation method provided by an embodiment of the present invention;

[0019] Figure 3is another flowchart of the product recommendation method provided by an embodiment of the present invention;

[0020] Figure 4 is a structural schematic diagram of a product recommendation device provided by an embodiment of the present invention;

[0021] Figure 5 It is a structural schematic diagram of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0022] Recommendation systems are widely used in e-commerce, content distribution, advertising and other fields, aiming to recommend products such as goods and services that users are interested in. In the prior art, recommendation systems recommend products to users in the following ways: recommendation methods based on user portraits, recommendation methods based on user social network information, and recommendation methods based on sequential decision making.

[0023] Among them, the recommendation method based on sequential decision-making has obvious limitations in the following aspects.

[0024] 1. Suboptimal recommendation results: Existing methods, such as upper bound optimization strategies or confidence interval optimization strategies, do not directly optimize the loss function of the problem itself. These strategies mainly focus on estimating the upper bound or confidence interval of the loss function, rather than directly minimizing the actual loss. This makes it difficult to achieve the optimal decision-making effect under limited budgets and resources, resulting in suboptimal recommendation results.

[0025] 2. Insufficient robustness due to parameter dependence: The performance of many sequential decision algorithms depends on given parameters, such as the distribution parameters of user preferences or the adjustment parameters in the algorithm. In some scenarios, such as cold start, these parameters are difficult to obtain accurately due to the lack of sufficient user data, making it difficult for the algorithm to quickly adapt and provide stable recommendation performance when facing new users or new products. Parameter dependence reduces the applicability and robustness of the algorithm in changing scenarios and increases the complexity and workload of algorithm tuning.

[0026] 3. Limitations of distribution assumptions: Existing technologies are usually based on specific assumptions about the distribution of user feedback, such as assuming that user ratings follow a normal distribution. However, in some application scenarios, these assumptions may not hold true because the actual distribution of user feedback may be unknown or not as expected. This limits the applicability and accuracy of the algorithm when dealing with new users or new products.

[0027] 4. Limitations of application scenarios: Most existing methods are mainly applicable to recommendation scenarios with continuous distribution, but are difficult to adapt to discrete distribution or more complex user behavior patterns. In addition, due to the need to optimize specific loss functions, the application of existing methods in a wider range of scenarios is limited, affecting their flexibility and efficiency.

[0028] A product recommendation method, device, electronic device and storage medium provided in the embodiment of the present invention, by drawing on the threshold multi-armed bandit problem, abstracting each candidate product into each arm of the threshold multi-armed bandit, optimizing the product recommendation method with the idea of ​​optimizing the threshold multi-armed bandit problem, and directly optimizing the loss function, can more accurately evaluate the potential value of each candidate product, so as to make more accurate recommendations; by using the large deviation theory to determine the candidate products recommended in each recommendation round in the first stage, it can converge faster within a limited number of recommendations, and because the candidate products recommended in each recommendation round are dynamically adjusted, it can respond quickly, improve the optimization efficiency of the method, facilitate obtaining more accurate expected returns of each candidate product, improve the overall optimization of the recommendation results, reduce the need for manual intervention, and reduce human resource consumption; because it does not rely on preset parameters, it can adaptively adjust its behavior when different user behaviors and product characteristics, thereby improving the flexibility and robustness of the algorithm; because it reduces the dependence on prior knowledge and reward distribution types, the algorithm of the present invention can be applied to a wider range of fields, especially in scenarios where cold start has limited data knowledge; and the present application can reduce computational complexity and reduce computer resource consumption.

[0029] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0030] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0031] Figure 1This is a flowchart of a product recommendation method provided by an embodiment of the present invention. This embodiment is applicable to the scenario of recommending products to new users. The method can be executed by a product recommendation device provided by an embodiment of the present invention, and the device can be implemented in software and / or hardware. In a specific embodiment, the device can be integrated into an electronic device, such as a computer, a server, etc. The following embodiments will be described by taking the device integrated into an electronic device as an example. Reference Figure 1 , the method may specifically include the following steps:

[0032] Step 101, in the first recommendation stage, based on the large deviation theory, the candidate products recommended to the user in each recommendation round are determined from the candidate products, the corresponding candidate products are recommended and the corresponding rewards are obtained. This step can improve the robustness and practicality of the product recommendation method as a whole by determining the products recommended in each recommendation round based on the large deviation theory without relying on pre-determined additional parameters, and can facilitate obtaining the sample rewards of each candidate product after the first recommendation stage and determining the expected rewards of each candidate product based on the sample returns of each candidate product.

[0033] Specifically, the above-mentioned user is preferably a new user, but may also be an old user.

[0034] Specifically, the candidate products mentioned above are preferably all products in the recommendation system that can be recommended to users, and may also be new products in the recommendation system.

[0035] Optionally, the first recommendation stage is the initial stage of recommending products to new users, or a stage after the initial stage.

[0036] Optionally, the above process of determining the candidate products to be recommended to users in each recommendation round from among various candidate products based on the large deviation theory includes: determining the candidate products to be recommended to users in each recommendation round from among various candidate products based on the large deviation theory according to empirical data, i.e., the number of times each candidate product is recommended to other users and the recommendation return obtained.

[0037] Specifically, the above process of obtaining the corresponding reward can be obtained based on the recommendation reward obtaining method in the prior art. For example, the recommendation reward can be calculated based on information such as the product recommended by the user in the corresponding round and the time of browsing the corresponding page.

[0038] Step 102, obtaining the returns of each recommended round of each candidate product at the end of the first recommendation stage to obtain the sample returns of each candidate product, and determining the expected return of each candidate product based on the sample returns of each candidate product. This step can help classify each candidate product into a recommendable product and a non-recommended product based on the expected return of each candidate product.

[0039] Specifically, the expected returns of the above-mentioned candidate products can be understood as the expected values ​​of the returns of the candidate products.

[0040] Optionally, the above process of determining the expected return of each candidate product based on the sample returns of each candidate product includes: for each candidate product, calculating the average of each sample return of the current candidate product to obtain the expected return of the current candidate product.

[0041] Specifically, the average value of the above-mentioned sample returns may be subjected to other processing in the prior art, such as calibration based on corresponding influencing factors, to obtain the expected return of the above-mentioned current candidate product. The influencing factors may be, for example, seasonal factors, weather factors, and major event factors of the user's area corresponding to the first recommendation stage.

[0042] In an optional specific example, the above process of calculating the average of the sample returns of the current candidate product for each candidate product to obtain the expected return of the current candidate product can be performed based on the following formula:

[0043]

[0044] in, represents the estimated value of the expected return of each candidate product, i represents each candidate product, m i represents the total amount of sample returns for each candidate product, X i,j It represents the sample return corresponding to the jth recommended round of the i-th candidate product.

[0045] Step 103, in the second recommendation phase, each candidate product is classified as a recommendable product or a non-recommended product based on the expected return, and the recommendable product is recommended to the user. Based on steps 101 and 102, this step avoids relying on pre-determined additional parameters by determining the products recommended to the user in each recommendation round from each candidate product based on the large deviation theory in the first recommendation phase; and obtaining the returns of each candidate product in each recommended round at the end of the first recommendation phase to obtain the sample returns of each candidate product, and then determining the expected return of each candidate product based on the sample returns of each candidate product, and further obtaining and recommending the recommendable product based on whether the expected return is greater than the expected return threshold, which can improve the applicability to different scenarios, especially the cold start scenario where the user is a new user, and improve the robustness of the recommendation, and obtain more accurate and reasonable recommendation results.

[0046] Optionally, the process of classifying each candidate product as a recommendable product or a non-recommended product based on expected returns includes:

[0047] For each candidate product, when the expected return of the current candidate product is greater than the expected return threshold of the current candidate product, the current candidate product is classified as a recommendable product; when the expected return of the previous candidate product is not greater than the expected return threshold, the current candidate product is classified as a non-recommended product.

[0048] Specifically, the expected return thresholds of the candidate products may be the same value or different values.

[0049] Specifically, the expected return threshold of each of the candidate products may be determined based on the cost of recommending each of the candidate products.

[0050] Optionally, the process of classifying each candidate product as a recommendable product or a non-recommended product based on expected returns includes:

[0051] The above-mentioned expected returns are subjected to other processing operations in the prior art, such as normalization, to obtain processed values, and then each candidate product is classified as a recommendable product or a non-recommended product based on a threshold value corresponding to the processed value.

[0052] In an optional specific example, each candidate product is represented as i (i∈{1,2,…,K}), and the reward obtained by recommending each product to the user is represented as X i , the expected return of product i is expressed as μ i , the expected return threshold of each candidate product is expressed as τ, Γ + (τ) represents the set of recommended products, Γ - (τ) represents the set of unrecommended products, then all candidate products can be classified into:

[0053] Γ + (τ)={i∈{1,2,…,K}:μ i >τ}, (1)

[0054] Γ - (τ)={i∈{1,2,…,K}:μ i ≤τ}, (2)

[0055] The following further describes a product recommendation method provided by another embodiment of the present invention. Figure 2 As shown, Figure 1 Step 101 in the method may include the following steps:

[0056] Step 1011, based on the large deviation theory, the current probability correlation number of each candidate product is estimated using the current number of recommended rounds and the current rewards of each candidate product.

[0057] Specifically, the current number of recommended rounds and the current rewards correspond to the number of recommended rounds of the corresponding candidate product and the rewards of each recommended round after completing the previous round of recommendations to the user and obtaining the corresponding rewards.

[0058] Specifically, the current returns of the above-mentioned candidate products can be understood as the combination of the returns of each recommended round of the corresponding candidate products.

[0059] Specifically, the current probability correlation number is a correlation number of the probability of classification error when classifying the corresponding candidate product based on the expected return of the corresponding candidate product determined based on the corresponding current existing return and the current number of recommended rounds.

[0060] Optionally, the process of estimating the current probability correlation number of the corresponding candidate product by using the current recommended round number and the current return of each candidate product based on the large deviation theory includes:

[0061] Based on the large deviation theory, the expected return threshold of each candidate product, the current number of recommended rounds and the current return are used to estimate the current probability correlation number of the corresponding candidate product.

[0062] Optionally, use m=(m1,m2,…,m K ) represents the number of times each candidate product has been recommended after any recommendation round is completed, and the average value of the existing returns after any recommendation round is expressed as When the expected return of each candidate product is expressed by the existing return after the corresponding recommendation round is completed, all candidate products can be classified to obtain the estimated classification results after each recommendation round is completed:

[0063]

[0064] in, It represents the set of recommended products in the estimated classification results after any recommendation round is completed. Represents the set of non-recommended products in the estimated classification results after any recommendation round is completed.

[0065] Optional, use P i (m) represents the probability of misclassification of each product, then:

[0066]

[0067] Based on the large deviation theory:

[0068]

[0069] Therefore, the probability of misclassification of the i-th candidate product can be expressed as:

[0070]

[0071] Optionally, the correlation number of the aforementioned classification error probability can be m i η i .

[0072] Specifically, the correlation number of the aforementioned classification error probability can also be -m i η i .

[0073] Optionally, the process of estimating the current probability correlation number of the corresponding candidate product by using the current recommended round number and the current return of each candidate product based on the large deviation theory includes:

[0074] Determine the function value of the rate function corresponding to the current recommendation round based on the current reward and the current number of recommendation rounds, where the rate function is the rate function in the large deviation theory; and

[0075] The current probability correlation number is determined based on the current recommended round number and the function value of the rate function corresponding to the current recommended round.

[0076] Optionally, the above process of determining the function value of the rate function corresponding to the current recommendation round based on the current existing returns and the current number of recommendation rounds includes: determining the function value of the rate function corresponding to the current recommendation round based on the current existing returns, the current number of recommendation rounds and the expected return threshold of each candidate product.

[0077] In a specific example, the process of determining the function value of the rate function corresponding to the current recommendation round based on the current existing rewards, the current number of recommended rounds and the expected reward threshold of each candidate product can be performed based on the following formula:

[0078]

[0079] in, represents the function value of the rate function; m i Indicates the number of times each candidate recommended product has been recommended after any recommendation round is completed; X i,j represents the reward corresponding to the j-th recommendation of candidate product i; θ represents an arbitrary value; τ represents the expected return threshold of each candidate product.

[0080] Optionally, the process of determining the current probability correlation number based on the current recommended round number and the function value of the rate function corresponding to the current recommended round includes: calculating

[0081] Specifically, based on the large deviation theory, the current number of recommended rounds, the current rewards, and the threshold corresponding to the aforementioned processed values ​​of each candidate product can be used to estimate the current probability correlation number of the corresponding candidate product.

[0082] Step 1012, based on the current probability correlation number of each candidate product, determine the recommended product for the current round from each candidate product.

[0083] Specifically, the current round of recommended products may include one candidate recommended product or multiple candidate recommended products.

[0084] Optionally, the process of determining the recommended product for the current round from each candidate product based on the current probability correlation number of each candidate product includes:

[0085] Determine a current function value of the loss function based on the current probability correlation number of each candidate product; and determine a recommended product for the current round from each candidate product based on the current function value of the loss function.

[0086] Specifically, in the estimated classification results after each of the above recommendation rounds, the total number of incorrectly classified candidate products L(m) can be expressed as:

[0087]

[0088] The above loss function can be expressed as:

[0089]

[0090] The estimator function of the above loss function can be expressed as:

[0091]

[0092] Optionally, the process of determining the recommended product for the current round from each candidate product based on the current function value of the loss function estimator includes:

[0093] Based on the change pattern of the current function value of the loss function estimator compared with the previous function value, the recommended product for the current round is determined from various candidate products.

[0094] Optionally, the process of determining the recommended product for the current round from each candidate product based on the current probability correlation number of each candidate product includes:

[0095] The candidate product with the smallest corresponding current probability correlation number among all candidate products is obtained to obtain the recommended product for the current round.

[0096] Specifically, from formula (8), we can see that for the probability correlation number The smaller the probability correlation number, the greater the probability of classification error of the corresponding candidate product. In order to optimize the loss function represented by formula (13), even if the loss function is minimized, the probability of classification error of each candidate product needs to be minimized. Therefore, it is necessary to optimize the function value of the loss function of the candidate product with the smallest probability correlation number, that is, to optimize the maximum value of the function value of the loss function corresponding to each candidate product.

[0097] Specifically, by directly optimizing the loss function rather than optimizing the upper bound and confidence interval of the loss function, it is possible to more accurately reflect and respond to actual user feedback, thereby obtaining a more effective product recommendation method.

[0098] Step 1013, recommend the current round recommended product in the current recommendation round, and obtain the corresponding reward to obtain the current round reward of the current round recommended product.

[0099] Step 1014: Update the current rewards of the corresponding candidate products based on the current round rewards, and update the current recommended round number of the recommended products in the current round.

[0100] Optionally, the process of updating the current existing returns of the corresponding candidate products based on the current round returns includes: adding the current round returns of each current round recommended product to the set of current existing returns of each current round recommended product.

[0101] Step 1015, restart the execution based on the large deviation theory, use the current recommended rounds and current returns of each candidate product to estimate the current probability correlation number of the corresponding candidate product, until the sum of the current recommended rounds of each candidate product is equal to the first value.

[0102] Specifically, the process of restarting the execution based on the large deviation theory, using the current number of recommended rounds and the current returns of each candidate product to estimate the current probability correlation number of the corresponding candidate product until the sum of the current number of recommended rounds of each candidate product is equal to the first value includes: repeatedly executing steps 1011-1014 until the sum of the current number of recommended rounds is equal to the first value.

[0103] Optionally, when the sum of the current recommended rounds of each candidate product is equal to a first value, the first recommendation phase ends.

[0104] Optionally, the first recommendation phase ends when the value of the loss function is minimized.

[0105] Specifically, steps 1011 to 1014 may be repeatedly performed until the sum of the current recommended rounds is equal to the first value.

[0106] Specifically, the first recommendation stage may also be ended when the value of the loss function is less than a preset loss function value threshold.

[0107] The embodiment of the present invention can provide a more direct and effective recommendation decision guide by directly minimizing the loss function, especially in a cold start scenario where user feedback data is scarce, and can obtain a more accurate and reasonable recommendation result.

[0108] The following further describes a product recommendation method provided by another embodiment of the present invention. Figure 3 As shown, the following steps may be included:

[0109] Step 301, in the initialization stage, directly recommend each candidate product of the second number of rounds to the user and obtain corresponding rewards to obtain the initialization sample rewards of each candidate product.

[0110] Specifically, the initialization phase may be understood as a phase of initially recommending each candidate product to a newly added user, or may be understood as a phase of initially recommending each newly added candidate product to a user.

[0111] Specifically, the second value is smaller than the first value.

[0112] Optionally, the sum of the second value and the total number of rounds of recommending each candidate product in the first recommendation stage is equal to the first value.

[0113] Step 302, in the first recommendation stage, based on the large deviation theory and using the initialization sample return, the candidate products recommended to the user in each recommendation round are determined from the candidate products, the corresponding candidate products are recommended and the corresponding returns are obtained.

[0114] Optionally, the process of determining the candidate products recommended to the user in each recommendation round from each candidate product by using the initialization sample return based on the large deviation theory includes:

[0115] Based on the large deviation theory, the current probability correlation number of each candidate product is estimated by using the current returns of each candidate product including the initialization sample returns and the current recommended rounds number of the initialization stage recommendation rounds.

[0116] Step 303: Obtain the returns of each recommended round of each candidate product at the end of the first recommendation phase to obtain a sample return of each candidate product, and determine the expected return of each candidate product based on the sample return of each candidate product.

[0117] Step 304, in the second recommendation stage, each candidate product is classified as a recommendable product or a non-recommended product based on the expected return, and the recommendable product is recommended to the user.

[0118] The embodiment of the present invention directly recommends the second numerical number of candidate recommended products in the initialization stage before the first recommendation stage, and obtains accurate user feedback information, which is conducive to efficiently and accurately obtaining the expected return of each candidate product, so as to make product recommendations more accurately and reasonably in the second recommendation stage.

[0119] Figure 4 is a structural diagram of a product recommendation device provided by an embodiment of the present invention, and the device is suitable for executing the product recommendation method provided by an embodiment of the present invention. Figure 4 As shown, the device may specifically include:

[0120] The first-stage recommendation module 401 is used to determine the candidate products recommended to the user in each recommendation round from the candidate products in the first recommendation stage based on the large deviation theory, recommend the corresponding candidate products and obtain the corresponding rewards. This module can improve the robustness and practicality of the product recommendation method as a whole by determining the products recommended in each recommendation round based on the large deviation theory without relying on pre-determined additional parameters, and can facilitate obtaining the sample rewards of each candidate product after the first recommendation stage and determining the expected rewards of each candidate product based on the sample returns of each candidate product.

[0121] Optionally, the first-stage recommendation module 401 can be specifically used to estimate the current probability correlation number of the corresponding candidate product based on the large deviation theory by using the current recommended round number and the current existing return of each candidate product; the current recommended round number and the current existing return correspond to the recommended round number and the return of each recommended round of the corresponding candidate product after completing the previous round of recommendation to the user and obtaining the corresponding return, and the current probability correlation number is the correlation number of the probability of classification error when classifying the corresponding candidate product based on the expected return of the corresponding candidate product determined based on the corresponding current existing return;

[0122] Based on the current probability correlation numbers of each candidate product, determine the recommended product for the current round from each candidate product;

[0123] Recommend the current round recommended product in the current recommendation round, and obtain the corresponding reward to obtain the current round reward of the current round recommended product;

[0124] Update the current return of the corresponding candidate product based on the current round return, and update the current recommended round number of the recommended product in the current round; and

[0125] Restart the execution based on the large deviation theory, use the current recommended rounds and current returns of each candidate product to estimate the current probability correlation number of the corresponding candidate product, until the sum of the current recommended rounds of each candidate product is equal to the first value.

[0126] Optionally, the first-stage recommendation module 401 can be specifically used to determine a function value of a rate function corresponding to a current recommendation round based on the current existing rewards and the current number of recommendation rounds, where the rate function is a rate function in the large deviation theory; and

[0127] The current probability correlation number is determined based on the current recommended round number and the function value of the rate function corresponding to the current recommended round.

[0128] Optionally, the first-stage recommendation module 401 can be specifically used to determine a current function value of the loss function based on the current probability correlation number of each candidate product; and

[0129] Based on the current function value of the loss function, the recommended product for the current round is determined from each candidate product. Optionally, the first stage recommendation module 401 can be specifically used to obtain the candidate product with the smallest corresponding current probability correlation number from each candidate product to obtain the recommended product for the current round.

[0130] The expected return acquisition module 402 is used to obtain the returns of each recommended round of each candidate product at the end of the first recommendation stage to obtain the sample returns of each candidate product, and determine the expected return of each candidate product based on the sample returns of each candidate product. This module can help classify each candidate product into a recommendable product and a non-recommended product based on the expected return of each candidate product.

[0131] Optionally, the expected return acquisition module 402 can be specifically used to calculate, for each candidate product, an average of the sample returns of the current candidate product to obtain the expected return of the current candidate product.

[0132] The second-stage recommendation module 403 is used to classify each candidate product as a recommendable product or a non-recommended product based on the expected return, and recommend the recommendable product to the user. This module is combined with modules 401 and 402, and avoids relying on pre-determined additional parameters by determining the products recommended to the user in each recommendation round from each candidate product based on the large deviation theory in the first recommendation stage; and obtains the returns of each candidate product in each recommended round at the end of the first recommendation stage to obtain the sample returns of each candidate product, and then determines the expected return of each candidate product based on the sample returns of each candidate product, and further obtains and recommends the recommendable product based on whether the expected return is greater than the expected return threshold, which can improve the applicability to different scenarios, especially the cold start scenario where the user is a new user, and improves the robustness of the recommendation, and obtains more accurate and reasonable recommendation results.

[0133] Optionally, the second-stage recommendation module 403 can be specifically used to classify, for each candidate product, the current candidate product as a recommendable product when the expected return of the current candidate product is greater than the expected return threshold of the current candidate product; and classify the current candidate product as a non-recommended product when the expected return of the previous candidate product is not greater than the expected return threshold.

[0134] Optionally, the product recommendation device provided by the embodiment of the present invention further includes: an initialization module, used to directly recommend each candidate product of the second numerical round to the user and obtain corresponding rewards to obtain initialization sample rewards of each candidate product.

[0135] Optionally, the aforementioned first-stage recommendation module 401 can be specifically used to determine the candidate products recommended to the user in each recommendation round from among the candidate products using the initialization sample returns based on the large deviation theory.

[0136] Those skilled in the art can clearly understand that for the convenience and simplicity of description, only the division of the above-mentioned functional modules is used as an example for illustration. In actual applications, the above-mentioned functions can be assigned to different functional modules as needed, that is, the internal structure of the device is divided into different functional modules to complete all or part of the functions described above. The specific working process of the functional modules described above can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here.

[0137] An embodiment of the present invention further provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the product recommendation method provided in any of the above embodiments is implemented.

[0138] An embodiment of the present invention further provides a computer-readable medium having a computer program stored thereon, and when the program is executed by a processor, the product recommendation method provided in any of the above embodiments is implemented.

[0139] An embodiment of the present invention further provides a computer program product, including a computer program, which, when executed by a processor, implements a product recommendation method as described in any one of the embodiments of the present invention.

[0140] Reference below Figure 5 , which shows a schematic diagram of the structure of a computer system 500 of an electronic device suitable for implementing an embodiment of the present invention. Figure 5 The electronic device shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present invention.

[0141] like Figure 5As shown, the computer system 500 includes a central processing unit (CPU) 501, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 502 or a program loaded from a storage part 508 into a random access memory (RAM) 503. In the RAM 503, various programs and data required for the operation of the system 500 are also stored. The CPU 501, the ROM 502, and the RAM 503 are connected to each other via a bus 504. An input / output (I / O) interface 505 is also connected to the bus 504.

[0142] The following components are connected to the I / O interface 505: an input section 506 including a keyboard, a mouse, etc.; an output section 507 including a cathode ray tube (CRT), a liquid crystal display (LCD), etc., and a speaker, etc.; a storage section 508 including a hard disk, etc.; and a communication section 509 including a network interface card such as a LAN card, a modem, etc. The communication section 509 performs communication processing via a network such as the Internet. A drive 510 is also connected to the I / O interface 505 as needed. A removable medium 511, such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc., is installed on the drive 510 as needed, so that a computer program read therefrom is installed into the storage section 508 as needed.

[0143] In particular, according to the embodiments disclosed in the present invention, the process described above with reference to the flowchart can be implemented as a computer software program. For example, the embodiments disclosed in the present invention include a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program includes a program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from the network through the communication part 509, and / or installed from the removable medium 511. When the computer program is executed by the central processing unit (CPU) 501, the above-mentioned functions defined in the system of the present invention are executed.

[0144] It should be noted that the computer-readable medium shown in the present invention may be a computer-readable signal medium or a computer-readable storage medium or any combination of the above two. The computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present invention, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in combination with an instruction execution system, device or device. In the present invention, a computer-readable signal medium may include a data signal propagated in a baseband or as part of a carrier wave, which carries a computer-readable program code. This propagated data signal may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. Computer-readable signal media may also be any computer-readable medium other than computer-readable storage media, which may send, propagate or transmit a program for use by or in conjunction with an instruction execution system, apparatus or device. The program code contained on the computer-readable medium may be transmitted using any appropriate medium, including but not limited to: wireless, wire, optical cable, RF, etc., or any suitable combination of the above.

[0145] The flow chart and block diagram in the accompanying drawings illustrate the possible architecture, function and operation of the system, method and computer program product according to various embodiments of the present invention. In this regard, each box in the flow chart or block diagram can represent a module, a program segment, or a part of a code, and the above-mentioned module, program segment, or a part of a code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram or flow chart, and the combination of the boxes in the block diagram or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0146] The modules and / or units involved in the embodiments of the present invention may be implemented in software or hardware. The modules and / or units described may also be set in a processor. For example, it may be described as: a processor includes a first-stage recommendation module, an expected return acquisition module, and a second-stage recommendation module. The names of these modules do not, in some cases, constitute limitations on the modules themselves.

[0147] As another aspect, the present invention also provides a computer-readable medium, which may be included in the device described in the above embodiment; or it may exist independently and not be assembled into the device. The above computer-readable medium carries one or more programs, and when the above one or more programs are executed by a device, the device implements: in the first recommendation stage, based on the large deviation theory, the candidate products recommended to the user in each recommendation round are determined from each candidate product, the corresponding candidate products are recommended and the corresponding rewards are obtained; the rewards of each recommended round of each candidate product at the end of the first recommendation stage are obtained to obtain the sample rewards of each candidate product, and the expected rewards of each candidate product are determined based on the sample rewards of each candidate product; and in the second recommendation stage, each candidate product is classified as a recommendable product or a non-recommended product based on the expected return, and the recommendable product is recommended to the user.

[0148] The above specific implementations do not constitute a limitation on the protection scope of the present invention. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions may occur depending on design requirements and other factors. Any modification, equivalent substitution and improvement made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A product recommendation method, It is characterized in that include: In the first recommendation stage, based on the large deviation theory, the candidate products recommended to users in each recommendation round are determined from the candidate products, and the corresponding candidate products are recommended and corresponding rewards are obtained; Obtaining the returns of each recommended round of each candidate product at the end of the first recommendation stage to obtain sample returns of each candidate product, and determining the expected returns of each candidate product based on the sample returns of each candidate product; as well as In the second recommendation stage, each candidate product is classified as a recommendable product or a non-recommended product based on the expected return, and the recommendable product is recommended to the user.

2. The product recommendation method according to claim 1, It is characterized in that The method of determining the candidate products recommended to the user in each recommendation round from the candidate products based on the large deviation theory includes: Based on the large deviation theory, the current probability correlation number of the corresponding candidate product is estimated by using the current recommended round number and the current existing return of each candidate product; the current recommended round number and the current existing return correspond to the recommended round number and the return of each recommended round of the corresponding candidate product after completing the previous round of recommendation to the user and obtaining the corresponding return, and the current probability correlation number is the correlation number of the probability of classification error when classifying the corresponding candidate product based on the expected return of the corresponding candidate product determined based on the corresponding current existing return; Determine a recommended product for the current round from among the candidate products based on the current probability correlation numbers of the candidate products; Recommending the current round recommended product in the current recommendation round, and obtaining a corresponding reward to obtain the current round reward of the current round recommended product; updating the currently available returns of the corresponding candidate products based on the current round returns, and updating the currently recommended round number of the recommended products in the current round; and Restart the execution based on the large deviation theory, and use the current number of recommended rounds and current returns of each candidate product to estimate the current probability correlation number of the corresponding candidate product, until the sum of the current number of recommended rounds of each candidate product is equal to the first value.

3. The product recommendation method according to claim 2, It is characterized in that In an initialization phase before the first recommendation phase, the product recommendation method further includes: Directly recommending each of the candidate products for a second number of rounds to the user and obtaining corresponding rewards to obtain initialization sample rewards for each of the candidate products; The method of determining the candidate products recommended to the user in each recommendation round from the candidate products based on the large deviation theory includes: Based on the large deviation theory, the candidate products recommended to the user in each recommendation round are determined from the candidate products according to the number of initialization recommendation rounds of the candidate products and the initialization sample returns.

4. The product recommendation method according to claim 2, It is characterized in that Based on the large deviation theory, the current probability correlation number of the corresponding candidate product is estimated by using the current recommended round number and the current return of each candidate product, including: Determine a function value of a rate function corresponding to a current recommendation round based on the current existing reward and the current number of recommendation rounds, wherein the rate function is a rate function in a large deviation theory; and The current probability correlation number is determined based on the current recommended round number and the function value of the rate function corresponding to the current recommended round.

5. The product recommendation method according to claim 2, It is characterized in that The step of determining a recommended product for the current round from among the candidate products based on the current probability correlation numbers of the candidate products includes: Determining a current function value of a loss function based on the current probability correlation number of each candidate product; and Based on the current function value of the loss function, the recommended product for the current round is determined from the candidate products.

6. The product recommendation method according to claim 2, It is characterized in that The step of determining a recommended product for the current round from among the candidate products based on the current probability correlation numbers of the candidate products includes: The candidate product with the smallest corresponding current probability correlation number among the candidate products is obtained to obtain the recommended product for the current round.

7. The product recommendation method according to claim 1, It is characterized in that The determining the expected return of each candidate product based on the sample return of each candidate product includes: For each candidate product, calculate the average of each sample return of the current candidate product to obtain the expected return of the current candidate product; The classifying each candidate product as a recommendable product or a non-recommended product based on the expected return includes: For each candidate product, when the expected return of the current candidate product is greater than the expected return threshold of the current candidate product, the current candidate product is classified as a recommendable product; when the expected return of the previous candidate product is not greater than the expected return threshold, the current candidate product is classified as a non-recommended product.

8. A product recommendation device, It is characterized in that include: The first-stage recommendation module is used to determine the candidate products recommended to users in each recommendation round from various candidate products in the first recommendation stage based on the large deviation theory, recommend the corresponding candidate products and obtain corresponding rewards; An expected return acquisition module, used to acquire the returns of each recommended round of each candidate product at the end of the first recommendation stage to obtain sample returns of each candidate product, and determine the expected return of each candidate product based on the sample returns of each candidate product; as well as The second stage recommendation module is used to classify the candidate products into recommendable products or non-recommended products based on the expected return, and recommend the recommendable products to the user.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, It is characterized in that When the processor executes the program, the product recommendation method as described in any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium having a computer program stored thereon, It is characterized in that When the program is executed by a processor, the product recommendation method as described in any one of claims 1 to 7 is implemented.