Article recommendation method and device

Abstract

The invention provides an article recommendation method and device, and the method comprises the steps: determining a first preference probability of a first user set for an SKU of a to-be-recommended article; the first user set comprises users who have interaction behaviors with the SKU of the to-be-recommended article; determining a second preference probability of a second user set for the SKU of the to-be-recommended article; the second user set comprises users who do not interact with the SKU of the to-be-recommended article; and determining a target user based on the first preference probability and the second preference probability, and recommending the to-be-recommended article to the target user. According to the embodiment, the item recommendation efficiency and accuracy can be improved.

Description

Technical Field

[0001] This invention relates to the field of information recommendation technology, and more specifically, to a method and apparatus for recommending items. Background Technology

[0002] In fields with big data, such as e-commerce, finance, and telecommunications, targeted recommendations are a common and effective marketing method. Taking e-commerce as an example, targeted marketing typically uses historical data on whether users who have interacted with a target product ultimately purchased it to predict potential buyers.

[0003] However, for new products and long-tail products, the positive sample size in historical interaction data is often too small to meet the need to predict future potential users based on historical interaction data. Existing technologies typically select historical interaction data of users who have interacted with the target product on related products to predict potential users of the target product. However, because the reference factor is singular and the reference priority of other related product data cannot be distinguished during prediction, the accuracy and efficiency of recommendations are lacking. Summary of the Invention

[0004] In view of this, the present disclosure provides at least one method, apparatus, electronic device, and storage medium for recommending items, which can improve the efficiency and accuracy of item recommendations.

[0005] In a first aspect, embodiments of this disclosure provide a method for recommending items, including:

[0006] Determine the first preference probability of the first user set for the SKU of the item to be recommended; the first user set includes users who have interacted with the SKU of the item to be recommended;

[0007] Determine the second preference probability of the second user set for the SKU of the item to be recommended; the second user set includes users who have not interacted with the SKU of the item to be recommended.

[0008] Based on the first preference probability and the second preference probability, the target user is determined, and the item to be recommended is recommended to the target user.

[0009] Optionally, a pre-trained diffusion prediction model is used to determine the second preference probability of the second user set for the SKU of the item to be recommended;

[0010] The diffusion prediction model is obtained through the following steps:

[0011] A sample SKU is extracted from the target SKU, and users who have interacted with the target SKU are extracted to obtain the first sample user set; the target SKU is different from the SKU of the item to be recommended.

[0012] Users who have interacted with the sampled SKU in the first sampled user set are removed to obtain a second sampled user set. Based on the sampled SKU and the second user set, a sampling sample pair is constructed.

[0013] Preference labels are added to the sampled pairs to determine the training sample set;

[0014] The diffusion prediction model is trained based on the training sample set.

[0015] Optionally, the target SKU and the SKU of the item to be recommended have a relationship in at least one dimension.

[0016] Optionally, after obtaining the sampled SKUs from the target SKUs, the process also includes:

[0017] Calculate the coverage of the sampled SKUs and determine whether the coverage is greater than or equal to a preset threshold;

[0018] If not, adjust the sampling rules and extract the sampled SKUs from the target SKUs according to the adjusted sampling rules.

[0019] Optionally, the diffusion prediction model is trained based on the training sample set, including:

[0020] Obtain the indicator values ​​for each sampled SKU across multiple dimensions; the multiple dimensions include at least one of the attribute dimension, industry dimension, and user dimension.

[0021] Based on the index value, calculate the content similarity value between the target SKU (excluding the sampled SKU) and the sampled SKU, and use a collaborative filtering algorithm to calculate the collaborative similarity value between the target SKU (excluding the sampled SKU) and the sampled SKU.

[0022] Based on the content similarity value and the collaborative similarity value, the sample features of the training sample set are obtained; based on the training sample set and the sample features, the diffusion prediction model is trained.

[0023] Optionally, determining the target user based on the first preference probability and the second preference probability includes:

[0024] The first user set is divided into multiple first user segments, and the second user set is divided into multiple second user segments;

[0025] The target user is determined based on the first preference probability corresponding to each of the first user segments and the second preference probability corresponding to each of the second user segments.

[0026] Optionally, the target user is determined based on the first preference probability corresponding to each of the first user segments and the second preference probability corresponding to each of the second user segments, including:

[0027] Calculate the evaluation metrics for each SKU of the item to be recommended for each of the first user segments and the second user segments;

[0028] Based on the calculation results of the evaluation indicators, each of the first user segments and the second user segments is sorted, and the target users are selected according to the sorting results.

[0029] Secondly, embodiments of this disclosure provide an item recommendation device, comprising:

[0030] The first determining module is used to determine the first preference probability of a first user set for the SKU of the item to be recommended; the first user set includes users who have interacted with the SKU of the item to be recommended.

[0031] The second determining module is used to determine the second preference probability of the second user set for the SKU of the item to be recommended; the second user set includes users who have not interacted with the SKU of the item to be recommended.

[0032] The recommendation module is used to determine the target user based on the first preference probability and the second preference probability, and recommend the item to be recommended to the target user.

[0033] Thirdly, embodiments of this disclosure also provide an electronic device, including: a processor, a memory, and a bus, wherein the memory stores machine-readable instructions executable by the processor, and when the computer device is running, the processor communicates with the memory via the bus, and when the machine-readable instructions are executed by the processor, the steps of the first aspect above, or any optional implementation of the first aspect, are performed.

[0034] Fourthly, embodiments of this disclosure also provide a computer-readable storage medium storing a computer program that, when executed by a processor, performs the steps of the first aspect or any optional implementation thereof.

[0035] In any of the above aspects or any implementation thereof, a first preference probability for a first user set regarding the SKU of the item to be recommended is determined; the first user set includes users who have interacted with the SKU of the item to be recommended; a second preference probability for a second user set regarding the SKU of the item to be recommended is determined; the second user set includes users who have not interacted with the SKU of the item to be recommended; based on the first preference probability and the second preference probability, a target user is determined, and the item to be recommended is recommended to the target user. Thus, by predicting the preference probability of the SKU of the item to be recommended from users who have interacted with it and users who have not interacted with it, the target user for the item to be recommended is determined, increasing the reference dimension when determining the target user, thereby improving the accuracy of item recommendation.

[0036] In a further embodiment of this disclosure, when determining the preference probability of the second user set for the SKU of the item to be recommended, a pre-trained diffusion prediction model can be used. During training the diffusion prediction model, a first sampled user set is obtained by extracting sampled SKUs from the target SKU and extracting users who have interacted with the target SKU; the target SKU is different from the SKU of the item to be recommended. Users who have interacted with the sampled SKU in the first sampled user set are removed to obtain the second sampled user set. Based on the sampled SKU and the second user set, sampled sample pairs are constructed. Preference labels are added to the sampled sample pairs to determine the training sample set. This improves the model's generalization ability, thus ensuring the accuracy and efficiency of item recommendations when facing items with insufficient positive samples in historical interaction data.

[0037] The effects of the recommended devices, electronic equipment, and storage media are described in the description of the recommended methods for the above-mentioned items, and will not be repeated here. Attached Figure Description

[0038] To more clearly illustrate the technical solutions of the embodiments of this disclosure, the accompanying drawings used in the embodiments will be briefly described below. These drawings are incorporated in and constitute a part of this specification. They illustrate embodiments conforming to this disclosure and, together with the specification, serve to explain the technical solutions of this disclosure. It should be understood that the following drawings only show some embodiments of this disclosure and should not be considered as limiting the scope. Those skilled in the art can obtain other related drawings based on these drawings without creative effort.

[0039] Figure 1 A flowchart of an item recommendation method provided by an embodiment of this disclosure is shown;

[0040] Figure 2 This illustration shows a flowchart of a method for recommending items provided in this disclosure, in which a high-potential prediction model and a diffusion prediction model are used to jointly determine the target user.

[0041] Figure 3 A schematic diagram of an item recommendation device provided in an embodiment of this disclosure is shown;

[0042] Figure 4 An exemplary system architecture in which embodiments of this disclosure can be applied is shown;

[0043] Figure 5 A schematic diagram of the structure of a computer system used to implement the embodiments of the present disclosure is shown. Detailed Implementation

[0044] To make the objectives, technical solutions, and advantages of the embodiments of this disclosure clearer, the technical solutions of the embodiments of this disclosure will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this disclosure, and not all of them. The components of the embodiments of this disclosure described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this disclosure provided in the accompanying drawings is not intended to limit the scope of the claimed disclosure, but merely represents selected embodiments of this disclosure. All other embodiments obtained by those skilled in the art based on the embodiments of this disclosure without inventive effort are within the scope of protection of this disclosure.

[0045] It should be noted that the collection, use, storage, sharing and transfer of user personal information involved in the technical solution of the present invention all comply with the provisions of relevant laws and regulations, and require notification to users and obtaining their consent or authorization. When applicable, user personal information is subjected to de-identification and / or anonymization and / or encryption technical processing.

[0046] Research has found that in fields with big data, such as e-commerce, finance, and telecommunications, targeted recommendation is a common and effective marketing method. Taking e-commerce as an example, targeted marketing typically predicts potential buyers of a target product based on historical data showing whether users who have interacted with the product ultimately purchased it. However, for items with a very small positive sample size in historical interaction data, such as newly listed products, traditional recommendation methods suffer from significant shortcomings in accuracy.

[0047] However, existing technologies have been optimized for recommending items without a sufficient number of positive samples. Typically, these technologies use historical interaction data of users who have interacted with the target product on related products as a reference. This approach maximizes the use of existing data to predict potential users for the target product. However, this method relies on a relatively singular reference dimension and fails to prioritize data from other related products when predicting potential users for the target product, resulting in shortcomings in both accuracy and efficiency of item recommendations.

[0048] Based on the above research, this disclosure provides an item recommendation method that can determine the target users of the recommended item by predicting the preference probabilities of users who have interacted with the SKU of the item to be recommended and users who have not interacted with the SKU of the item to be recommended. This increases the reference dimension when determining the target users, thereby improving the accuracy of item recommendation. Furthermore, when determining the preference probability of users who have not interacted with the SKU of the item to be recommended, a pre-trained diffusion prediction model can be used. During the training of the diffusion prediction model, a sample SKU is extracted from the target SKU, and users who have interacted with the target SKU are extracted to obtain a first sampled user set. The target SKU is different from the SKU of the item to be recommended. Users who have interacted with the sampled SKU in the first sampled user set are removed to obtain a second sampled user set. Based on the sampled SKU and the second user set, sampled sample pairs are constructed. Preference labels are added to the sampled sample pairs to determine the training sample set. This improves the model's generalization ability, thus ensuring the accuracy and efficiency of item recommendations when faced with items where the number of positive samples in historical interaction data is insufficient.

[0049] The above problems and solutions are the result of the inventor's practice and careful research. The discovery process of the above problems and the solutions proposed for the above problems should be considered as the inventor's contribution to this disclosure.

[0050] The technical solutions of this disclosure will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this disclosure, and not all of them. The components of this disclosure described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this disclosure provided in the accompanying drawings is not intended to limit the scope of the claimed disclosure, but merely to illustrate selected embodiments of this disclosure. All other embodiments obtained by those skilled in the art based on the embodiments of this disclosure without inventive effort are within the scope of protection of this disclosure.

[0051] It should be noted that similar labels and letters in the following figures indicate similar items. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.

[0052] To facilitate understanding of this embodiment, a detailed description of the item recommendation method disclosed in this disclosure is provided first. The execution entity of the item recommendation method provided in this disclosure is generally a computer device with a certain computing capability. This computer device may include, for example, a terminal device, a server, or other processing devices. The terminal device may be a user equipment (UE), mobile device, user terminal, terminal, cellular phone, cordless phone, personal digital assistant (PDA), handheld device, computing device, in-vehicle device, wearable device, etc. In some possible implementations, the item recommendation method can be implemented by a processor calling computer-readable instructions stored in memory.

[0053] See Figure 1 The diagram shown is a flowchart of the item recommendation method provided in this embodiment. It should be noted that the specific implementation methods described below are merely illustrative examples from the e-commerce retail field and do not constitute an improper limitation of the present invention. The method includes steps S101-S103, wherein:

[0054] S101: Determine the first preference probability of the first user set for the SKU of the item to be recommended; the first user set includes users who have interacted with the SKU of the item to be recommended.

[0055] In this step, a first set of users who have not interacted with the SKU of the item to be recommended can be determined based on historical data, and then the preference probability of the first set of users for the SKU of the item to be recommended can be determined.

[0056] In this embodiment, the preference probability can be a binary question such as whether a user will purchase the corresponding product after browsing the SKU on an e-commerce platform. In this case, the preference probability value can be 1 or 0 (1 indicates that the user is predicted to purchase the recommended item, and 0 indicates that the user will not purchase the recommended item). In one feasible implementation, the preference probability value can also be a specific numerical value, such as predicting that one user has a 30% probability of purchasing the recommended item, and another user has an 85% probability of purchasing the recommended item.

[0057] In this embodiment of the disclosure, the preference probability of the first user set for the SKU of the recommended item can be determined by a pre-trained high-potential prediction model; or, the preference probability of the first user set for the SKU of the recommended item can be predicted directly based on historical data.

[0058] In one possible implementation, training samples for training the high-potential prediction model can be obtained first. The format of the training samples can be "user code + SKU code + feature set + preference label", where the feature set can be obtained based on historical data over a period of time. For example, user interaction data with SKUs within the past 180 days can be selected and summarized into the feature set in the training samples, such as user preferences at the SKU granularity, brand granularity, price range granularity, industry, and attribute levels. The preference label here can be set according to needs to indicate whether the user purchased the SKU within 7 days after interacting with it. If so, the preference label can be set to 1; otherwise, it can be set to 0. It should be noted that the length of the time period after the user interacts with the SKU and decides whether to purchase it can be set according to needs. The above description of 7 days is only an example and does not constitute an improper limitation of the present invention.

[0059] In this embodiment of the disclosure, after obtaining the training samples of the high-potential prediction model, the user code, SKU code and feature set can be used as the input features of the high-potential prediction model, and the preference note can be used as the training target of the high-potential prediction model to train the high-potential prediction model.

[0060] In this embodiment of the disclosure, the high-potential prediction model predicts the probability that a set of users who have interacted with a target SKU within a certain period will purchase the interacting SKU in the future. For example, if 10,000 people have browsed a certain SKU1 within 180 days, the model predicts the probability that these 10,000 users will purchase that SKU1 within the next 7 days.

[0061] S102: Determine the second preference probability of the second user set for the SKU of the item to be recommended; the second user set includes users who have not interacted with the SKU of the item to be recommended.

[0062] In this embodiment of the disclosure, for long-tail products and new products, due to the lack of sample size and historical data, it is impossible to use a high-potential prediction model to predict the probability of users purchasing these items. Existing technologies generally determine the target users of a target SKU by referring to information such as the preferred price range and preferred product categories of competing products of the target SKU by users who have interacted with it. However, this method suffers from low accuracy due to considering only one factor. Therefore, this embodiment of the disclosure proposes, in addition to predicting the probability of purchase of a SKU by users who have interacted with it, to further predict the probability of purchase of a SKU by users who have not interacted with it.

[0063] In practice, the second preference probability of the second user set for the SKU of the item to be recommended can be determined by a pre-trained diffusion prediction model, or the second preference probability of the second user set for the SKU of the item to be recommended can be predicted directly by historical data.

[0064] In this embodiment of the disclosure, the diffusion prediction model is obtained through the following steps: a sample SKU is extracted from the target SKU, and users who have interacted with the target SKU are extracted to obtain a first sampled user set; the target SKU is different from the SKU of the item to be recommended; users who have interacted with the sampled SKU in the first sampled user set are removed to obtain a second sampled user set; based on the sampled SKU and the second user set, a sampling sample pair is constructed; preference labels are added to the sampling sample pair to determine the training sample set; and the diffusion prediction model is trained based on the training sample set.

[0065] It should be noted that the target SKU here can be an SKU that has at least one dimension of relationship with the SKU of the item to be recommended. Specifically, the relationship between the target SKU and the SKU of the item to be recommended can be that they belong to the same industry. For example, when the SKU of the item to be recommended is a refrigerator of a certain brand, the target SKU can be any brand or model of refrigerator within the refrigerator industry. In addition, the relationship between the target SKU and the SKU of the item to be recommended can also be that they have the same specific attribute value. For example, if the SKU of the item to be recommended is a gold double-door refrigerator of brand A, then the target SKU can be a refrigerator of another brand, or a refrigerator of another appearance under brand A. Alternatively, the relationship between the target SKU and the SKU of the item to be recommended can also be that they have similar or related functions, such as refrigerators and freezers, refrigerators and refrigerator disinfectants, refrigerator storage boxes, etc.

[0066] In this embodiment of the disclosure, the diffusion prediction model can be trained using a generalized sample pair approach. Specifically, sampling can be performed on the target SKU by obtaining historical orders from the past year and calculating the average daily sales of each SKU. Considering the new product launch date, the average daily sales can be calculated by comparing the order sales from the past year with the order span in days (last order date - first order date). Then, data cleaning can be performed to remove data that does not conform to the norm. For example, data cleaning could involve determining the minimum price within the industry. For instance, assuming the target SKU's industry is refrigerators, SKUs priced below 100 yuan are not actually refrigerators and are considered invalid data; these can be filtered out. The average daily sales of items priced above this level can then be calculated. When calculating the average daily sales, the number of orders can be used as a dimension for segmentation, such as the number and percentage of SKUs with 1-5 daily sales, or the number and percentage of SKUs with 5-10 daily sales. It should be noted that the threshold price for data cleaning here can be obtained from statistical analysis of historical data from e-commerce platforms, or it can be determined based on industry experience.

[0067] In this embodiment, after calculating the average daily sales, the sampling rules can be determined. Sampling is then performed in each segment of the target SKU based on the number of SKUs and their proportion. For example, for SKUs with an average daily sales of more than 10 units, since the proportion is low, all can be sampled; for SKUs with an average daily sales of 6-10 units, the number of units is also relatively small, and the proportion is low, so 50% is sampled; for SKUs with an average daily sales of 4-5 units, 30% is sampled; for SKUs with an average daily sales of 2-3 units, 10% is sampled; and for SKUs with an average daily sales of 1 unit, 1% is sampled. It should be noted that the sampling rules can be set as needed during implementation, but it is necessary to ensure that the sampling rules can cover the characteristics of all products across the industry. The sampling method exemplified above can cover the vast majority of brand attributes and prices within the industry, as well as products with different sales volumes. Based on the Pareto principle (the top 20% of best-selling items account for 80% of the total sales of a category, while the remaining 80% of items account for 20%), there will be relatively few top-selling items in each category. If random sampling were used, there would be very few or even none of these top-selling items, which would skew the characteristics of the sampled items. Therefore, we use daily average sales for segmented sampling. In addition, we use daily average sales to take into account the product's launch time. If we used total sales, newly launched items would have relatively low total sales and would be categorized into mid-tier or long-tail items. If we then sampled based on the sampling ratio, the probability of sampling new products would be relatively small.

[0068] In this embodiment, after sampling SKUs are obtained from the target SKUs, the coverage of the sampled SKUs can be verified. Specifically, the coverage of the sampled SKUs in terms of brand, price, and key attributes (e.g., in the refrigerator industry, key attributes could be door type, color, height, cooling capacity, power consumption level, etc.) can be verified. When each coverage is greater than or equal to a preset threshold, the sampling is considered successful. For example, brand coverage can be calculated by comparing the number of brands of the sampled SKUs to the number of brands of all active SKUs in the industry; price segment coverage can be calculated by comparing the number of price segments of the sampled SKUs to the number of price segments of all active SKUs in the industry; and key attribute coverage can be calculated by comparing the number of key attribute values ​​of the sampled SKUs to the number of attribute values ​​of all active SKUs in the industry. When brand coverage is above 85%, price segment coverage is 100%, and key attribute value coverage is above 85%, the sampling is considered successful. If the sampled SKUs do not meet the above conditions, the sampling is considered a failure. The sampling rules can be adjusted for resampling until the coverage of the sampled SKUs meets the requirements. It should be noted that the dimensions and thresholds for the coverage verification can be set according to requirements. The above is merely an example of one possible implementation method and does not constitute an undue limitation of the present invention.

[0069] In this embodiment of the disclosure, after sampling the target SKU to obtain the sampled SKU, users can be sampled to obtain a first sampled user set. For example, the sampling rule for users here can be to randomly select 10% of users who have browsed the refrigerator industry within the past 30 days.

[0070] In practical implementation, since the diffusion prediction model predicts the probability of preference for a SKU among users who have not interacted with it, it is necessary to remove users who have interacted with the sampled SKU from the first sampled user set to obtain the second sampled user set. For example, if a user has interacted with 10 SKUs in the refrigerator industry (SKU1, SKU2, ..., SKU10), and SKU1 is the sampled SKU, then that user is not in the sample pair. After removing the sampled users, the second sampled user set and the sampled SKUs can be paired to form sample pairs. The format of the sample pair can be "second sampled user code + sampled SKU code". Furthermore, since the above sampling involves users who have browsed the refrigerator industry within the past 30 days, data from multiple time periods can be merged, avoiding major promotional periods and ensuring the data spans both peak and off-peak seasons across industries. This will give the final sample pairs stronger generalization capabilities. In addition, preference labels can be added to the sample pairs after sampling is completed; for example, if the second sampled user browsed the sampled SKU from December 16 to January 14, whether they purchased the sampled SKU from January 15 to January 21 is recorded as 1, otherwise 0; if they browsed the sampled SKU from March 16 to April 14, whether they purchased the sampled SKU from April 15 to April 21 is recorded as 1, otherwise 0; if they browsed the sampled SKU from July 16 to July 14, whether they purchased the sampled SKU from July 15 to July 21 is recorded as 1, otherwise 0.

[0071] In this embodiment of the disclosure, for the multiple sample pairs obtained, random sampling can be performed again. Here, the sampling rules can only refer to the volume of sample data required for training the diffusion prediction model. The training sample format at this time can be "second sampling user code + sampling SKU code + preference label + time period".

[0072] In this embodiment of the disclosure, after obtaining sample pairs for training the diffusion prediction model, a feature set can be constructed for them to train the diffusion prediction model, including: obtaining the indicator values ​​of each sampled SKU under multiple dimensions; the multiple dimensions include at least one of attribute dimension, industry dimension, and user dimension; based on the indicator values, calculating the content similarity values ​​between the target SKU (excluding the sampled SKU) and the sampled SKU, and using a collaborative filtering algorithm to calculate the collaborative similarity values ​​between the target SKU (excluding the sampled SKU) and the sampled SKU; obtaining the sample features of the training sample set based on the content similarity values ​​and the collaborative similarity values; and training the diffusion prediction model based on the training sample set and the sample features.

[0073] In practical implementation, taking the refrigerator industry as an example, when constructing a feature set for sample pairs, it can include attribute dimensions, such as SKUA brand, price range, new product launch time, capacity, door type, and refrigeration type; it can also include sales dimensions, such as average daily sales and positive review rate; it can also include industry dimensions, such as page views (PV), number of users adding items to their cart, and number of searches; and it can also include user dimensions, such as user purchasing power and user browsing price trajectory.

[0074] In this embodiment of the disclosure, after determining the above features, a collaborative filtering algorithm can be used to calculate the collaborative similarity values ​​between the target SKU (excluding the sampled SKU) and the sampled SKU. A higher score indicates a higher similarity between the two SKUs. The calculation formula can be the ratio of the number of simultaneously interested users of the two SKUs to the square root of (number of interested users of the baseline SKU * number of interested users of the related SKU). Furthermore, an industry vector benchmark can be used to construct an SKU vector using the attribute features of the SKU. Then, cosine similarity can be used to calculate the pairwise content similarity values ​​between the target SKU (excluding the sampled SKU) and the sampled SKU. For example, the refrigerator industry vector benchmark is {price, brand, door type, color, cooling type, capacity…}, and SKUA is {900, Brand A, double door, silver, inverter, 200L…}. Each dimension in the vector is replaced with the attention level of that attribute value, and then cosine similarity is calculated. It should be noted that other similarity calculation methods can also be used when calculating content similarity values, and this disclosure does not limit this.

[0075] In this embodiment of the disclosure, after calculating the collaborative similarity value and the content similarity value, the similarity features can be processed. For example, the preference value, collaborative filtering value, collaborative similarity prediction value (SKU preference value * collaborative similarity value with SKUA) and content similarity prediction value (SKU preference value * content similarity value with SKUA) of the top five SKUs with the highest similarity to SKUA can be extracted as part of the features.

[0076] S103: Based on the first preference probability and the second preference probability, determine the target user and recommend the item to be recommended to the target user.

[0077] In this step, after predicting the first preference probability of users who have interacted with the product and the second preference probability of users who have not interacted with the product, the target users can be identified based on the two preference probabilities, and the product can be recommended to the target users.

[0078] In this embodiment of the disclosure, determining a target user based on the first preference probability and the second preference probability includes: dividing the first user set into multiple first user segments and dividing the second user set into multiple second user segments; determining a target user based on the first preference probability corresponding to each first user segment and the second preference probability corresponding to each second user segment.

[0079] In practice, the first user set and the second user set can be divided into multiple segments, for example, each segment can be 1,000 people. Then, the preference probability corresponding to each segment can be calculated, and users in the segment with higher preference probability can be directly selected as target users.

[0080] In this embodiment of the disclosure, determining the target user based on the first preference probability corresponding to each first user segment and the second preference probability corresponding to each second user segment includes: calculating the evaluation index for each first user segment and the second user segment for the SKU of the item to be recommended; sorting each first user segment and the second user segment according to the calculation results of the evaluation index; and selecting the target user according to the sorting results.

[0081] In practice, evaluation metrics such as SKU conversion rate of the recommended items can be calculated for each user segment. After obtaining the evaluation metrics, user segments whose evaluation metrics exceed the preset threshold can be directly selected as target users. Alternatively, they can be sorted in descending order and the user segments with the highest ranking can be selected as target users.

[0082] One possible implementation, such as Figure 2 As shown, a high-potential prediction model can be used to predict the probability of preference for a specific SKU among users who have interacted with it, while a diffusion prediction model can be used to predict the probability of preference for a specific SKU among users who have not interacted with it. By combining the prediction results of both models, the target users for the recommended item can be determined. The diffusion prediction model is trained by constructing generalized sample pairs and building a feature set for each pair. This improves the model's generalization ability, ensuring the accuracy and efficiency of item recommendations even when dealing with items with insufficient positive samples in historical interaction data.

[0083] According to a second aspect of the embodiments of this disclosure, such as Figure 3 As shown, an item recommendation device 300 is provided, comprising:

[0084] The first determining module 301 is used to determine the first preference probability of a first user set for the SKU of the item to be recommended; the first user set includes users who have interacted with the SKU of the item to be recommended.

[0085] The second determining module 302 is used to determine the second preference probability of the second user set for the SKU of the item to be recommended; the second user set includes users who have not interacted with the SKU of the item to be recommended;

[0086] The recommendation module 303 is used to determine the target user based on the first preference probability and the second preference probability, and recommend the item to be recommended to the target user.

[0087] Optionally, the second determining module 302 is specifically used for:

[0088] A sample SKU is extracted from the target SKU, and users who have interacted with the target SKU are extracted to obtain the first sample user set; the target SKU is different from the SKU of the item to be recommended.

[0089] Users who have interacted with the sampled SKU in the first sampled user set are removed to obtain a second sampled user set. Based on the sampled SKU and the second user set, a sampling sample pair is constructed.

[0090] Preference labels are added to the sampled pairs to determine the training sample set;

[0091] The diffusion prediction model is trained based on the training sample set.

[0092] Optionally, the target SKU and the SKU of the item to be recommended have a relationship in at least one dimension.

[0093] Optionally, the second determining module 302 is further configured to:

[0094] Calculate the coverage of the sampled SKUs and determine whether the coverage is greater than or equal to a preset threshold;

[0095] If not, adjust the sampling rules and extract the sampled SKUs from the target SKUs according to the adjusted sampling rules.

[0096] Optionally, the second determining module 302 is specifically used for:

[0097] Obtain the indicator values ​​for each sampled SKU across multiple dimensions; the multiple dimensions include at least one of the attribute dimension, industry dimension, and user dimension.

[0098] Based on the index value, calculate the content similarity value between the target SKU (excluding the sampled SKU) and the sampled SKU, and use a collaborative filtering algorithm to calculate the collaborative similarity value between the target SKU (excluding the sampled SKU) and the sampled SKU.

[0099] Based on the content similarity value and the collaborative similarity value, the sample features of the training sample set are obtained; based on the training sample set and the sample features, the diffusion prediction model is trained.

[0100] Optionally, module 303 is specifically used for:

[0101] The first user set is divided into multiple first user segments, and the second user set is divided into multiple second user segments;

[0102] The target user is determined based on the first preference probability corresponding to each of the first user segments and the second preference probability corresponding to each of the second user segments.

[0103] Optionally, module 303 is specifically used for:

[0104] Calculate the evaluation metrics for each SKU of the item to be recommended for each of the first user segments and the second user segments;

[0105] Based on the calculation results of the evaluation indicators, each of the first user segments and the second user segments is sorted, and the target users are selected according to the sorting results.

[0106] According to a third aspect of the present disclosure, an electronic device for recommending items is provided, comprising: one or more processors; and a storage device for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the method provided in the first aspect of the present invention.

[0107] According to a fourth aspect of the present disclosure, a computer-readable medium is provided having a computer program stored thereon, which, when executed by a processor, implements the method provided in the first aspect of the present invention.

[0108] Figure 4 An exemplary system architecture 400 is shown that can be applied to the item recommendation method or item recommendation apparatus implemented in this disclosure.

[0109] like Figure 4As shown, system architecture 400 may include terminal devices 401, 402, and 403, a network 404, and a server 405. Network 404 serves as the medium for providing communication links between terminal devices 401, 402, and 403 and server 405. Network 404 may include various connection types, such as wired or wireless communication links, or fiber optic cables, etc.

[0110] Users can use terminal devices 401, 402, and 403 to interact with server 405 via network 404 to receive or send messages, etc. Various communication client applications can be installed on terminal devices 401, 402, and 403, such as shopping applications, web browser applications, search applications, instant messaging tools, email clients, social media platform software, etc. (for example only).

[0111] Terminal devices 401, 402, and 403 can be various electronic devices with displays and web browsing capabilities, including but not limited to smartphones, tablets, laptops, and desktop computers.

[0112] Server 405 can be a server that provides various services, such as a backend management server that supports shopping websites browsed by users using terminal devices 401, 402, and 403 (for example only). The backend management server can process received item recommendation requests and send the processing results (for example only) back to the terminal device.

[0113] It should be noted that the item recommendation method provided in this embodiment of the invention is generally executed by server 405, and correspondingly, the item recommendation device is generally located in server 405. The item recommendation method provided in this embodiment of the invention can also be executed by terminal devices 401, 402, and 403, and correspondingly, the item recommendation device can be located in terminal devices 401, 402, and 403.

[0114] It should be understood that Figure 4 The number of terminal devices, networks, and servers shown is merely illustrative. Depending on implementation needs, any number of terminal devices, networks, and servers can be included.

[0115] The following is for reference. Figure 5 It shows a schematic diagram of the structure of a computer system 500 suitable for implementing a terminal device of the present invention. Figure 5 The terminal device shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of the present invention.

[0116] like Figure 5As shown, the computer system 500 includes a central processing unit (CPU) 501, which can perform various appropriate actions and processes based on programs stored in read-only memory (ROM) 502 or programs loaded from storage section 508 into random access memory (RAM) 503. The RAM 503 also stores various programs and data required for the operation of the system 500. The CPU 501, ROM 502, and RAM 503 are interconnected via a bus 504. An input / output (I / O) interface 505 is also connected to the bus 504.

[0117] The following components are connected to I / O interface 505: an input section 506 including a keyboard, mouse, etc.; an output section 507 including a cathode ray tube (CRT), liquid crystal display (LCD), etc., and speakers, 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, modem, etc. The communication section 509 performs communication processing via a network such as the Internet. A drive 510 is also connected to I / O interface 505 as needed. A removable medium 511, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed on drive 510 as needed so that computer programs read from it can be installed into storage section 508 as needed.

[0118] In particular, according to the embodiments disclosed in this invention, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments disclosed in this invention include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via communication section 509, and / or installed from removable medium 511. When the computer program is executed by central processing unit (CPU) 501, it performs the functions defined above in the system of this invention.

[0119] It should be noted that the computer-readable medium shown in this invention can be a computer-readable signal medium or a computer-readable storage medium, or any combination thereof. A computer-readable storage medium can be, for example,—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this invention, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In this invention, a computer-readable signal medium can include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. Computer-readable signal media can also be any computer-readable medium other than computer-readable storage media, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to: wireless, wire, optical fiber, RF, etc., or any suitable combination thereof.

[0120] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present invention. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram or flowchart, and combinations of blocks in a block diagram or flowchart, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0121] The modules described in the embodiments of the present invention can be implemented in software or hardware. The described modules can also be housed in a processor. For example, a processor may include a first determining module, a second determining module, and a recommending module. The names of these modules do not necessarily limit the specific module itself. For instance, the recommending module could be described as "a module that determines a target user based on the first preference probability and the second preference probability, and recommends the item to be recommended to the target user."

[0122] In another aspect, the present invention also provides a computer-readable medium, which may be included in the device described in the above embodiments; or it may exist independently and not assembled into the device. The computer-readable medium carries one or more programs, which, when executed by the device, implement the following method: determining a first preference probability of a first user set for a SKU of an item to be recommended; the first user set includes users who have interacted with the SKU of the item to be recommended;

[0123] Determine the second preference probability of the second user set for the SKU of the item to be recommended; the second user set includes users who have not interacted with the SKU of the item to be recommended.

[0124] Based on the first preference probability and the second preference probability, the target user is determined, and the item to be recommended is recommended to the target user.

[0125] Finally, it should be noted that the above-described embodiments are merely specific implementations of this disclosure, used to illustrate the technical solutions of this disclosure, and not to limit it. The protection scope of this disclosure is not limited thereto. Although this disclosure has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that any person skilled in the art can still modify or easily conceive of changes to the technical solutions described in the foregoing embodiments, or make equivalent substitutions for some of the technical features, within the scope of the technology disclosed in this disclosure. Such modifications, changes, or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this disclosure, and should all be covered within the protection scope of this disclosure. Therefore, the protection scope of this disclosure should be determined by the protection scope of the claims.

Claims

1. A method for recommending items, characterized in that, include: Determine the first preference probability of the first user set for the SKU of the item to be recommended; The first user set includes users who have interacted with the SKU of the item to be recommended; Determine the second preference probability of the second user set for the SKU of the item to be recommended; the second user set includes users who have not interacted with the SKU of the item to be recommended. Based on the first preference probability and the second preference probability, the target user is determined, and the item to be recommended is recommended to the target user.

2. The method according to claim 1, characterized in that, Using a pre-trained diffusion prediction model, determine the second preference probability of the second user set for the SKU of the item to be recommended; The diffusion prediction model is obtained through the following steps: A sample SKU is extracted from the target SKU, and users who have interacted with the target SKU are extracted to obtain the first sample user set; the target SKU is different from the SKU of the item to be recommended. Users who have interacted with the sampled SKU in the first sampled user set are removed to obtain a second sampled user set. Based on the sampled SKU and the second user set, a sampling sample pair is constructed. Preference labels are added to the sampled pairs to determine the training sample set; The diffusion prediction model is trained based on the training sample set.

3. The method according to claim 2, characterized in that, The target SKU and the SKU of the item to be recommended have a correlation in at least one dimension.

4. The method according to claim 2, characterized in that, After obtaining the sampled SKUs from the target SKUs, the following is also included: Calculate the coverage of the sampled SKUs and determine whether the coverage is greater than or equal to a preset threshold; If not, adjust the sampling rules and extract the sampled SKUs from the target SKUs according to the adjusted sampling rules.

5. The method according to claim 2, characterized in that, Based on the training sample set, the diffusion prediction model is trained, including: Obtain the indicator values ​​for each sampled SKU across multiple dimensions; the multiple dimensions include at least one of the attribute dimension, industry dimension, and user dimension. Based on the index value, calculate the content similarity value between the target SKU (excluding the sampled SKU) and the sampled SKU, and use a collaborative filtering algorithm to calculate the collaborative similarity value between the target SKU (excluding the sampled SKU) and the sampled SKU. Based on the content similarity value and the collaborative similarity value, the sample features of the training sample set are obtained; based on the training sample set and the sample features, the diffusion prediction model is trained.

6. The method according to claim 1, characterized in that, Based on the first preference probability and the second preference probability, the target user is determined, including: The first user set is divided into multiple first user segments, and the second user set is divided into multiple second user segments; The target user is determined based on the first preference probability corresponding to each of the first user segments and the second preference probability corresponding to each of the second user segments.

7. The method according to claim 6, characterized in that, Based on the first preference probability corresponding to each of the first user segments and the second preference probability corresponding to each of the second user segments, the target user is determined, including: Calculate the evaluation metrics for each SKU of the item to be recommended for each of the first user segments and the second user segments; Based on the calculation results of the evaluation indicators, each of the first user segments and the second user segments is sorted, and the target users are selected according to the sorting results.

8. An item recommendation device, characterized in that, include: The first determining module is used to determine the first preference probability of the first user set for the SKU of the item to be recommended; The first user set includes users who have interacted with the SKU of the item to be recommended. The second determining module is used to determine the second preference probability of the second user set for the SKU of the item to be recommended; the second user set includes users who have not interacted with the SKU of the item to be recommended. The recommendation module is used to determine the target user based on the first preference probability and the second preference probability, and recommend the item to be recommended to the target user.

9. An electronic device, characterized in that, include: One or more processors; Storage device for storing one or more programs. When the one or more programs are executed by the one or more processors, the one or more processors implement the method as described in any one of claims 1-7.

10. A computer-readable medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the method as described in any one of claims 1-7.

Images