Recommendation method, device and equipment based on user interest and computer program product

Abstract

The invention discloses a recommendation method, device and equipment based on user interests and a computer program product. The method comprises the steps of obtaining attribute data and historical behavior data of a user; calculating a long-term interest score and an overall interest score of the user at each interest point according to the historical behavior data; according to the long-term interest score of each interest point and the user data, utilizing a multi-target model based on a user interest efficiency tower to predict a short-term interest score of each interest point; and according to the short-term interest score and the long-term interest score of each interest point and the overall interest score, determining a fused interest score of each interest point, so as to recommend contents to the user. According to the method, the network structure of the user interest efficiency tower is introduced into the multi-target model, so that the memory ability of the model for long-term interest characteristics is enhanced; by fusing the long and short-term interest points, the interest of the user is ensured to conform to the short-term change rule and to be converged to the long-term interest distribution, the accuracy of interest estimation is improved, and the accuracy of recommended content is further improved.

Description

Technical Field

[0001] The present application relates to the technical field of interest recommendation, and in particular to a recommendation method, device and equipment based on user interests, and a computer program product. Background Art

[0002] In the recommendation system, the prediction of users' multiple interests is the focus of the recommendation system. For example, male users' interests are mainly concentrated in sports, military, games, etc., while female users' interests are mainly concentrated in beauty, entertainment, food, etc. At the same time, male users also have different preferences for the same interest point. For example, senior football fans have a significantly higher preference for sports than ordinary male users. Therefore, the core task of the recommendation system's multi-interest prediction is to estimate the user's preference for various fields based on the user's interests.

[0003] At present, the multi-interest prediction scheme of the recommendation system is mainly implemented using a multi-task deep learning model. The input of the multi-task deep learning model is user-related features, and the output is the model prediction score of the user in various interest areas.

[0004] However, in actual data analysis, it was found that users' interests will gradually stabilize over time and will not change significantly. The interest prediction based on the existing multi-task deep learning model is more based on users' short-term behavior and lacks long-term behavior. Therefore, the actual prediction results are not accurate enough. Summary of the invention

[0005] The embodiments of the present application provide a recommendation method, apparatus, device, and computer program product based on user interests to improve the accuracy of interest estimation and thereby improve the accuracy of recommended content.

[0006] The present application embodiment adopts the following technical solutions:

[0007] In a first aspect, an embodiment of the present application provides a recommendation method based on user interests, and the recommendation method based on user interests includes:

[0008] Acquire user data, wherein the user data includes attribute data and historical behavior data of the user;

[0009] Calculating the user's long-term interest score at each point of interest based on the user's historical behavior data and calculating the user's overall interest score based on the user's historical behavior data;

[0010] According to the long-term interest score of the user at each point of interest and the user data, a multi-objective model based on a user interest efficiency tower is used to predict the short-term interest score of the user at each point of interest;

[0011] Determining the integrated interest score of the user at each point of interest according to the short-term interest score and the long-term interest score and the overall interest score of the user at each point of interest;

[0012] Recommend content to users based on their integrated interest scores at various interest points.

[0013] Optionally, calculating the user's long-term interest score at each point of interest based on the user's historical behavior data includes:

[0014] Counting the user's click data at each point of interest and the exposure data of each point of interest based on the user's historical behavior data;

[0015] The user's long-term interest score at each point of interest is calculated based on the user's click data at each point of interest and the exposure data at each point of interest.

[0016] Optionally, calculating the overall interest score of the user according to the historical behavior data of the user includes:

[0017] Counting the user's overall click data and overall exposure data based on the user's historical behavior data;

[0018] The user's overall interest score is calculated based on the user's overall click data and overall exposure data.

[0019] Optionally, determining the integrated interest score of the user at each interest point according to the short-term interest score, the long-term interest score and the overall interest score of the user at each interest point includes:

[0020] Compare the short-term interest score with the long-term interest score of each interest point to obtain a first comparison result;

[0021] Compare the long-term interest score of each interest point with the overall interest score to obtain a second comparison result;

[0022] The fusion interest score of the user at each point of interest is determined according to the first comparison result and the second comparison result.

[0023] Optionally, determining the fused interest score of the user at each point of interest according to the first comparison result and the second comparison result includes:

[0024] If the first comparison result is that the short-term interest score of the interest point is greater than the long-term interest score of the interest point, and the second comparison result is that the long-term interest score of the interest point is greater than the overall interest score, and the difference between the short-term interest score of the interest point and the long-term interest score of the interest point is not less than the difference between the long-term interest score of the interest point and the overall interest score, then the short-term interest score and the long-term interest score of the interest point are merged to obtain a merged interest score of the interest point;

[0025] If the first comparison result is that the difference between the short-term interest score of the point of interest and the long-term interest score of the point of interest fluctuates within a preset interval, the short-term interest score of the point of interest is used as the fusion interest score of the point of interest, and the preset interval is determined according to the difference between the long-term interest score of the point of interest and the overall interest score;

[0026] Otherwise, the long-term interest score of the interest point is used as the fusion interest score of the interest point.

[0027] Optionally, the recommending content to the user according to the integrated interest points of the user at various interest points includes:

[0028] Sort the user's integrated interest scores at each point of interest, and recommend content corresponding to the target point of interest to the user based on the sorting result; or,

[0029] The fused interest score of the user at each point of interest is compared with a preset score threshold to obtain a third comparison result, and content corresponding to the target point of interest is recommended to the user according to the third comparison result.

[0030] Optionally, the multi-objective model based on the user interest efficiency tower includes an expert network, a gating network and an interest efficiency tower, and the multi-objective model based on the user interest efficiency tower is trained in the following manner:

[0031] Obtain user sample data and the user's long-term interest score at each point of interest, wherein the user sample data includes the user's attribute data and historical behavior data;

[0032] Generating user sample feature data according to the user sample data, and inputting the user sample feature data into the expert network to obtain original output features of the expert network;

[0033] Inputting the user's long-term interest scores at each point of interest into the interest efficiency tower to obtain output features of the interest efficiency tower;

[0034] Fusing the original output features of the expert network and the output features of the interest efficiency tower to obtain fused features;

[0035] According to the fusion features, using the gating network corresponding to each point of interest to predict the user's short-term interest score at each point of interest;

[0036] The predicted loss value is calculated according to the user's short-term interest score at each point of interest, and the parameters of the multi-objective model based on the user interest efficiency tower are optimized using the predicted loss value to obtain a trained multi-objective model based on the user interest efficiency tower.

[0037] In a second aspect, an embodiment of the present application further provides a recommendation device based on user interests, the recommendation device based on user interests comprising:

[0038] An acquisition unit, used to acquire user data, wherein the user data includes attribute data and historical behavior data of the user;

[0039] a calculation unit, configured to calculate the user's long-term interest score at each point of interest based on the user's historical behavior data and to calculate the user's overall interest score based on the user's historical behavior data;

[0040] A prediction unit, configured to predict the user's short-term interest score at each point of interest according to the user's long-term interest score at each point of interest and the user data, using a multi-objective model based on a user interest efficiency tower;

[0041] A fusion unit, configured to determine a fusion interest score of the user at each interest point according to the short-term interest score and the long-term interest score and the overall interest score of the user at each interest point;

[0042] The recommendation unit is used to recommend content to users based on the integrated interest scores of users at various interest points.

[0043] In a third aspect, an embodiment of the present application further provides a device, including:

[0044] A processor; and a memory arranged to store computer executable instructions, which, when executed, cause the processor to perform any of the aforementioned recommendation methods based on user interests.

[0045] In a fourth aspect, an embodiment of the present application further provides a computer program product, including a computer program / instruction, which, when executed by a processor, implements any of the aforementioned user interest-based recommendation methods.

[0046] At least one of the above technical solutions adopted in the embodiments of the present application can achieve the following beneficial effects: the user interest-based recommendation method in the embodiments of the present application first obtains user data, which includes user attribute data and historical behavior data; then calculates the user's long-term interest score at each interest point based on the user's historical behavior data and calculates the user's overall interest score based on the user's historical behavior data; then, based on the user's long-term interest score at each interest point and the user data, a multi-objective model based on the user interest efficiency tower is used to predict the user's short-term interest score at each interest point; then, based on the user's short-term interest score, long-term interest score and overall interest score at each interest point, the user's fused interest score at each interest point is determined; finally, content is recommended to the user based on the user's fused interest score at each interest point. The user interest-based recommendation method in the embodiments of the present application introduces the network structure of the user interest efficiency tower into the multi-objective model, which strengthens the output layer's memory ability for long-term interest features; by fusing long-term and short-term interest scores, it is ensured that the user's interest conforms to the short-term change law and converges to the long-term interest distribution, thereby improving the accuracy of user interest estimation and thus improving the accuracy of recommended content. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0048] Figure 1 A flowchart of a recommendation method based on user interests in an embodiment of the present application;

[0049] Figure 2 This is a schematic diagram of the overall structure of a multi-objective model based on a user interest efficiency tower in an embodiment of the present application;

[0050] Figure 3 This is a schematic diagram of the structure of a recommendation device based on user interests in an embodiment of the present application;

[0051] Figure 4 This is a schematic diagram of the structure of a device in an embodiment of the present application. DETAILED DESCRIPTION

[0052] In order to make the purpose, technical solution and advantages of the present application clearer, the technical solution of the present application will be clearly and completely described below in combination with the specific embodiments of the present application and the corresponding drawings. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present application.

[0053] The technical solutions provided by various embodiments of the present application are described in detail below in conjunction with the accompanying drawings.

[0054] Existing interest estimation solutions based on multi-task deep learning models have the following main problems:

[0055] (1) The user's long-term stable historical behavior is not considered. In the existing multi-interest prediction scheme, most user behavior features are based on short-term (such as the past month) behavior features to predict interests. There are fewer historical behaviors, which leads to the overall interest prediction being more biased towards short-term interests. Generally, the overall interests of users tend to be stable or convergent. Therefore, the prediction results of the model based on short-term behavior training will deviate from the user's long-term stable interests.

[0056] (2) Long-term user behavior has not been fully strengthened. In the existing network structure, the features closer to the output layer are more important and can have a greater impact on the model results, while the features closer to the input layer have less information and have little impact on the results. Therefore, in the traditional multi-interest prediction scheme, even if the user's long-term behavior is added, as the model is passed layer by layer, the information that can be learned in the output layer of the model is often limited, resulting in less long-term behavior that can be learned.

[0057] Based on this, the present application embodiment provides a recommendation method based on user interests, such as Figure 1 As shown, a flowchart of a recommendation method based on user interests in an embodiment of the present application is provided, and the recommendation method based on user interests at least includes the following steps S110 to S150:

[0058] Step S110, obtaining user data, wherein the user data includes user attribute data and historical behavior data.

[0059] When recommending content based on user interests, it is necessary to first obtain user data, including basic attribute data and historical behavior data of the user. Basic attribute data may include, for example, data such as the user's gender and age, and historical behavior data may include, for example, data such as the user's historical click behavior and browsing behavior on the recommended content. Of course, the specific dimensions of user data to be obtained can be flexibly set by those skilled in the art according to actual needs, and no specific limitation is made here.

[0060] Step S120, calculating the long-term interest score of the user at each point of interest based on the historical behavior data of the user and calculating the overall interest score of the user based on the historical behavior data of the user.

[0061] After obtaining the user's historical behavior data, the user's historical behavior data can be statistically analyzed. On the one hand, the user's long-term interest score at each point of interest can be calculated by statistically analyzing the user's click behavior on the recommended content at each point of interest. On the other hand, the user's overall historical click behavior can be statistically analyzed to calculate the user's overall interest score.

[0062] The long-term interest score of a user at each interest point reflects whether the user has a relatively long-term and stable preference behavior at each interest point, which is mainly used to make up for the deviation of short-term interest score prediction in the future. The overall interest score of a user reflects the overall preference behavior of the user for all recommended content, which is mainly used to eliminate the influence of individual difference behavior in the future.

[0063] Step S130, predicting the user's short-term interest score at each point of interest according to the user's long-term interest score at each point of interest and the user data using a multi-objective model based on the user interest efficiency tower.

[0064] The embodiment of the present application has pre-trained a multi-objective model based on the user interest efficiency tower. The multi-objective model based on the user interest efficiency tower is based on the traditional multi-objective model, and adds a network structure of an interest efficiency tower. The interest efficiency tower mainly adopts a fully connected layer structure, and its function is to strengthen the multi-objective model's learning of user long-term interest characteristics.

[0065] By combining the user's long-term interest score at each interest point, the user's attribute data and behavior data, the multi-objective model based on the user interest efficiency tower can be used to predict the user's short-term interest score at each interest point.

[0066] It should be noted here that the reason why the score predicted by the multi-objective model based on the user interest efficiency tower is used as the user's short-term interest score is mainly because in the traditional user multi-interest prediction scheme based on the multi-task model, although the multi-task model can learn the user's interest probability, after layer-by-layer learning, the user's long-term interests that can be learned at the network output layer are very limited, that is, the user has more short-term interests and fewer long-term interests, which causes the user's interest probability to be more biased towards short-term interests. Therefore, the traditional multi-task model actually outputs the user's short-term interest score that lacks long-term behavior correction.

[0067] Furthermore, based on the traditional multi-task model, the embodiment of the present application further introduces an interest efficiency tower structure that can strengthen long-term interest features, and uses the input of long-term interest features to correct the short-term interest score output by the model, which essentially obtains a more accurate short-term interest score.

[0068] Step S140, determining the integrated interest score of the user at each interest point according to the short-term interest score, the long-term interest score and the overall interest score of the user at each interest point.

[0069] Based on the short-term interest score, long-term interest score and overall interest score of the user at each interest point obtained in the above steps, it is necessary to further integrate the long-term and short-term interest scores of the user at each interest point, and finally obtain the fused interest score of the user at each interest point, so that the final fused interest score can not only reflect that the user's interest always fluctuates around the long-term interest point, but also meet the user's short-term interest needs.

[0070] Step S150: recommending content to the user based on the user's integrated interest scores at various interest points.

[0071] After calculating the user's integrated interest score at each interest point, the integrated interest scores of all interest points can be comprehensively considered to recommend relevant content to the user. The specific recommendation method can be flexibly set by technicians in this field according to actual needs and is not specifically limited here.

[0072] The user interest-based recommendation method of the embodiment of the present application introduces the network structure of the user interest efficiency tower into the multi-objective model, strengthening the output layer's memory ability for long-term interest features; by fusing long-term and short-term interest scores, it ensures that the user's interests conform to both short-term change rules and converge to the long-term interest distribution, thereby improving the accuracy of user interest estimation and thereby improving the accuracy of recommended content.

[0073] In some embodiments of the present application, calculating the long-term interest score of the user at each point of interest based on the user's historical behavior data includes: counting the user's click data at each point of interest and the exposure data of each point of interest based on the user's historical behavior data; calculating the user's long-term interest score at each point of interest based on the user's click data at each point of interest and the exposure data of each point of interest.

[0074] For the calculation of the long-term interest score of the user at each interest point, the exposure and click volume of the recommended items under each interest point can be counted according to the user's historical behavior log, and the click rate of the user at each interest point can be calculated according to the exposure and click volume of the recommended items under each interest point. The click rate can represent the conversion efficiency, and the click rate of the user at each interest point is used as the long-term interest score of the user at each interest point. For example, user A has a total of 10 recommended sports items exposed under the sports interest, of which user A clicked on 5 recommended sports items, that is, the click rate of user A under the sports interest is 5 / 10=0.5.

[0075] By calculating the user's long-term interest score at each interest point, we can more accurately understand the user's long-term interests and preferences, and can serve as a basis for subsequent corrections to the short-term interest score, which in turn helps provide users with more accurate and personalized recommendations.

[0076] In some embodiments of the present application, calculating the user's overall interest score based on the user's historical behavior data includes: counting the user's overall click data and overall exposure data based on the user's historical behavior data; and calculating the user's overall interest score based on the user's overall click data and overall exposure data.

[0077] For the calculation of the user's overall interest score, the total exposure and total clicks of the user's recommended items can be counted based on the user's historical behavior log, that is, regardless of the specific type of recommended items, the user's overall click rate is calculated based on the total exposure and total clicks of the user's recommended items as the overall interest score. For example, if user A has cumulatively exposed 20 recommended items, of which user A clicked on 5 recommended items, then user A's overall interest score is 5 / 20=0.25.

[0078] In some embodiments of the present application, determining the user's integrated interest score at each point of interest based on the user's short-term interest score, long-term interest score and overall interest score at each point of interest includes: comparing the short-term interest score with the long-term interest score of each point of interest to obtain a first comparison result; comparing the long-term interest score with the overall interest score of each point of interest to obtain a second comparison result; and determining the user's integrated interest score at each point of interest based on the first comparison result and the second comparison result.

[0079] After calculating the user's long-term interest score and short-term interest score at each interest point and the user's overall interest score based on the aforementioned embodiments, the user's long-term and short-term interest scores can be compared according to certain comparison rules, and an appropriate fusion strategy can be adopted according to the comparison results to obtain the user's fused interest score at each interest point.

[0080] Specifically, the user's short-term interest score at each point of interest can be compared with the long-term interest score, and the user's long-term interest score at each point of interest can be compared with the overall interest score. The comparison results of the interest scores can determine whether the short-term interest score of each point of interest predicted by the model is significantly higher or lower than the long-term interest score, and then different fusion strategies can be adopted to calculate the user's fused interest score at each point of interest.

[0081] The above comparison and judgment method can fully consider the long-term interest score and short-term interest score of the user in each interest point, avoiding the situation where the short-term interest score predicted by the model is too high and too much content of short-term interest points is recommended, ignoring the long-term interest points of the user, and also avoiding the situation where the short-term interest score predicted by the model is too low and no recommendation is made.

[0082] In some embodiments of the present application, determining the fused interest score of the user at each point of interest based on the first comparison result and the second comparison result includes: if the first comparison result is that the short-term interest score of the point of interest is greater than the long-term interest score of the point of interest, and the second comparison result is that the long-term interest score of the point of interest is greater than the overall interest score, and the difference between the short-term interest score of the point of interest and the long-term interest score of the point of interest is not less than the difference between the long-term interest score of the point of interest and the overall interest score, then the short-term interest score and the long-term interest score of the point of interest are fused to obtain the fused interest score of the point of interest; if the first comparison result is that the difference between the short-term interest score of the point of interest and the long-term interest score of the point of interest fluctuates within a preset interval, then the short-term interest score of the point of interest is used as the fused interest score of the point of interest, and the preset interval is determined according to the difference between the long-term interest score of the point of interest and the overall interest score; otherwise, the long-term interest score of the point of interest is used as the fused interest score of the point of interest.

[0083] When determining the fusion score of each interest point based on the comparison result of the above embodiment, the fusion interest score Ctr in different situations can be determined in the following manner: i :

[0084]

[0085] Among them, Ctr i represents the user's long-term interest score for interest i, Ctr all Represents the user's overall interest score. Represents the user's short-term interest score for interest i.

[0086] The above formula is explained as follows: when the user's short-term interest score is significantly higher than the long-term interest score, in order to prevent the interest point from being biased by the short-term interest, the fusion mechanism of the average of the long-term and short-term interest scores can be used to calculate the fusion interest score. When the difference between the long-term and short-term interest scores fluctuates within a range, it means that the interest score fluctuates within a certain range. At this time, the short-term interest score can be directly used as the fusion interest score. When the short-term interest score is significantly lower than the long-term interest score, the long-term interest score can be used as the fusion interest score to ensure convergence to the user's long-term interest.

[0087] In general, if the short-term interest score output by the model is too high, the user's short-term interest will be weakened to avoid the recommendation results being all short-term results. For example, during the European Cup, a large number of football videos cannot be recommended to users, but long-term results should be recommended to ensure the diversity of recommendation results. On the contrary, when the short-term interest score is too low, recommendations are given to users based on the long-term interest score to avoid results that are not recommended to users due to the short-term interest score being too low, thereby achieving "awakening" of user interest.

[0088] In some embodiments of the present application, the recommending content to the user based on the user's integrated interest scores at various points of interest includes: sorting the user's integrated interest scores at various points of interest, and recommending content corresponding to the target point of interest to the user based on the sorting result; or, comparing the user's integrated interest scores at various points of interest with a preset score threshold to obtain a third comparison result, and recommending content corresponding to the target point of interest to the user based on the third comparison result.

[0089] The level of the fusion interest score reflects the user's interest level in each point of interest. Therefore, the user's fusion interest scores at each point of interest can be sorted in descending order. According to the sorting results, recommended items corresponding to the top points of interest can be recommended to the user. Recommended items corresponding to points of interest with fusion interest scores greater than a certain score threshold can also be recommended to the user. The number and frequency of recommended content can also be controlled according to the level of the fusion interest score.

[0090] The specific method of recommending items based on the integrated interest score can be flexibly configured by those skilled in the art according to the needs of the actual application scenario, and is not specifically limited here.

[0091] In some embodiments of the present application, the multi-objective model based on the user interest efficiency tower includes an expert network, a gating network and an interest efficiency tower, and the multi-objective model based on the user interest efficiency tower is trained in the following manner: obtaining user sample data and the user's long-term interest score at each interest point, the user sample data including the user's attribute data and historical behavior data; generating user sample feature data based on the user sample data, and inputting the user sample feature data into the expert network to obtain the original output feature of the expert network; inputting the user's long-term interest score at each interest point into the interest efficiency tower to obtain the output feature of the interest efficiency tower; fusing the original output feature of the expert network with the output feature of the interest efficiency tower to obtain the fused feature; based on the fused feature, using the gating network corresponding to each interest point to predict the user's short-term interest score at each interest point; calculating the predicted loss value based on the user's short-term interest score at each interest point, and using the predicted loss value to optimize the parameters of the multi-objective model based on the user interest efficiency tower to obtain a trained multi-objective model based on the user interest efficiency tower.

[0092] In the existing user multi-interest estimation scheme, most of them use MMOE (Multi-gate Mixture-of-Experts, multi-gate mixed expert network) deep learning network. The learning goal of the network is the probability of multiple interests of the user. The user's real-time training features are used as samples to finally achieve the probability estimation of multiple points of interest of the user. The MMOE network consists of an expert network and a gated network. The expert network uses a multi-layer neural network. Each expert network is used to learn user features respectively. The gated network is used to "score" the output of each expert network, and the expert's scores are comprehensively weighted. Finally, the weighted results of the experts are used as the predicted interest probability.

[0093] In the above scheme, although the MMOE deep neural network can learn the user interest probability, after layers of learning by the expert network, the user's long-term interests that can be learned in the network output layer are very limited, that is, the user's short-term interests are more and long-term interests are less, resulting in the user's interest probability being more biased towards short-term interests and the overall estimation convergence is poor.

[0094] Based on this, Figure 2 As shown, a schematic diagram of the overall structure of a multi-objective model based on a user interest efficiency tower in an embodiment of the present application is provided. Based on the MMOE network, the embodiment of the present application further introduces the network structure of the interest efficiency tower (Interest-Tower). The interest efficiency tower can be regarded as a fully connected layer, and its function is to strengthen the output layer's "memory" ability for long-term interests, that is, the closer to the output layer, the stronger the memory ability.

[0095] When training the multi-objective model based on the user interest tower efficiency tower (Interest-Tower MMOE, IT-MMOE), the input of the model is user attribute characteristics (such as gender, age, etc.), behavior characteristics (such as user click behavior, etc.), and the model output is the probability distribution of each user's interest point. Within the model, the number of gated networks can be set according to the number of outputted interest points, and the expert network uses a deep neural network. It is worth noting that the embodiment of the present application introduces the structure of interest efficiency towers in the expert network, the number of interest efficiency towers is consistent with the number of expert networks, the input of the interest efficiency tower is the statistical result of long-term interest points, and the outputted interest efficiency features are weightedly fused with the features outputted by the expert network, ensuring that the user's long-term interest can be further strengthened in the expert network.

[0096] The multi-objective model based on the user interest efficiency tower proposed in the embodiment of the present application strengthens the user's long-term behavior characteristics through the introduction of the interest efficiency tower, ensuring that the user's long-term interest characteristics can be learned by the model.

[0097] The present application embodiment also provides a recommendation device 300 based on user interests, such as Figure 3 As shown, a schematic diagram of the structure of a recommendation device based on user interests in an embodiment of the present application is provided. The recommendation device based on user interests 300 includes: an acquisition unit 310, a calculation unit 320, a prediction unit 330, a fusion unit 340 and a recommendation unit 350, wherein:

[0098] An acquisition unit 310 is used to acquire user data, wherein the user data includes attribute data and historical behavior data of the user;

[0099] A calculation unit 320, configured to calculate the user's long-term interest score at each point of interest based on the user's historical behavior data and to calculate the user's overall interest score based on the user's historical behavior data;

[0100] A prediction unit 330, configured to predict the short-term interest score of the user at each point of interest according to the long-term interest score of the user at each point of interest and the user data by using a multi-objective model based on a user interest efficiency tower;

[0101] A fusion unit 340, configured to determine a fusion interest score of the user at each interest point according to the short-term interest score and the long-term interest score and the overall interest score of the user at each interest point;

[0102] The recommendation unit 350 is used to recommend content to the user according to the integrated interest points of the user at various interest points.

[0103] In some embodiments of the present application, the calculation unit 320 is specifically used to: count the user's click data at each point of interest and the exposure data of each point of interest based on the user's historical behavior data; calculate the user's long-term interest score at each point of interest based on the user's click data at each point of interest and the exposure data of each point of interest.

[0104] In some embodiments of the present application, the calculation unit 320 is specifically used to: count the user's overall click data and overall exposure data based on the user's historical behavior data; calculate the user's overall interest score based on the user's overall click data and overall exposure data.

[0105] In some embodiments of the present application, the fusion unit 340 is specifically used to: compare the short-term interest score of each interest point with the long-term interest score to obtain a first comparison result; compare the long-term interest score of each interest point with the overall interest score to obtain a second comparison result; determine the user's fused interest score at each interest point based on the first comparison result and the second comparison result.

[0106] In some embodiments of the present application, the fusion unit 340 is specifically used to: if the first comparison result is that the short-term interest score of the interest point is greater than the long-term interest score of the interest point, and the second comparison result is that the long-term interest score of the interest point is greater than the overall interest score, and the difference between the short-term interest score of the interest point and the long-term interest score of the interest point is not less than the difference between the long-term interest score of the interest point and the overall interest score, then the short-term interest score and the long-term interest score of the interest point are fused to obtain the fused interest score of the interest point; if the first comparison result is that the difference between the short-term interest score of the interest point and the long-term interest score of the interest point fluctuates within a preset interval, then the short-term interest score of the interest point is used as the fused interest score of the interest point, and the preset interval is determined according to the difference between the long-term interest score of the interest point and the overall interest score; otherwise, the long-term interest score of the interest point is used as the fused interest score of the interest point.

[0107] In some embodiments of the present application, the recommendation unit 350 is specifically used to: sort the user's fused interest scores at various points of interest, and recommend content corresponding to the target point of interest to the user based on the sorting result; or, compare the user's fused interest scores at various points of interest with a preset score threshold to obtain a third comparison result, and recommend content corresponding to the target point of interest to the user based on the third comparison result.

[0108] In some embodiments of the present application, the multi-objective model based on the user interest efficiency tower includes an expert network, a gating network and an interest efficiency tower, and the multi-objective model based on the user interest efficiency tower is trained in the following manner: obtaining user sample data and the user's long-term interest score at each interest point, the user sample data including the user's attribute data and historical behavior data; generating user sample feature data based on the user sample data, and inputting the user sample feature data into the expert network to obtain the original output feature of the expert network; inputting the user's long-term interest score at each interest point into the interest efficiency tower to obtain the output feature of the interest efficiency tower; fusing the original output feature of the expert network with the output feature of the interest efficiency tower to obtain the fused feature; based on the fused feature, using the gating network corresponding to each interest point to predict the user's short-term interest score at each interest point; calculating the predicted loss value based on the user's short-term interest score at each interest point, and using the predicted loss value to optimize the parameters of the multi-objective model based on the user interest efficiency tower to obtain a trained multi-objective model based on the user interest efficiency tower.

[0109] It can be understood that the above-mentioned recommendation device based on user interests can implement the various steps of the recommendation method based on user interests provided in the aforementioned embodiments. The relevant explanations on the recommendation method based on user interests are applicable to the recommendation device based on user interests and will not be repeated here.

[0110] Figure 4 Schematic diagram of the structure of a device in the embodiment of the present application. Figure 4 As shown, the device includes one or more processors (or processing units), may further include one or more memories coupled to the processors, and may further include a communication module coupled to the processors.

[0111] The communication module can be used to communicate with other devices or apparatuses, such as the transmission or reception of data and / or signals. The communication module can have at least one communication module for communication. The communication module can include any interface necessary for communicating with other devices. Exemplarily, the communication module can be a transceiver, a circuit, a bus, a module, or other types of communication modules.

[0112] The processor may include, but is not limited to, at least one of the following: a general-purpose computer, a special-purpose computer, a microcontroller, a digital signal controller (DSP), or one or more of a controller-based multi-core controller architecture. The device may have multiple processors, such as application-specific integrated circuit chips, which are time-dependent and synchronized with a clock of a main processor.

[0113] The memory may include one or more non-volatile memories and one or more volatile memories. Examples of non-volatile memories include, but are not limited to, at least one of the following: read-only memory (ROM), erasable programmable read-only memory (EPROM), flash memory, hard disk, compact disc (CD), digital video disc (DVD), or other magnetic storage and / or optical storage. Examples of volatile memories include, but are not limited to, at least one of the following: random access memory (RAM), or other volatile memories that do not persist during the duration of a power outage.

[0114] The computer program includes computer executable instructions executed by an associated processor. The program can be stored in ROM. The processor can perform any suitable actions and processes by loading the program into RAM.

[0115] The possible implementation of the present application can be implemented by means of a program, so that the communication device can perform any process discussed in the above embodiments. The possible implementation of the present application can also be implemented by hardware or by a combination of software and hardware.

[0116] In some embodiments, the program may be tangibly contained in a computer-readable storage medium, which may be included in the device (such as in a memory) or other storage device accessible by the device. The program may be loaded from the computer-readable storage medium to the RAM for execution. The computer-readable storage medium may include any type of tangible non-volatile memory, such as ROM, EPROM, flash memory, hard disk, CD, DVD, etc.

[0117] The present application embodiment also provides a computer-readable storage medium, on which computer instructions or program codes are stored, and when the processor runs the instructions or the program codes, the processor executes the methods and functions involved in any of the above embodiments. Computer-readable media can be any tangible medium containing or storing programs for or related to instruction execution systems, devices or equipment. Computer-readable media can be computer-readable signal media or computer-readable storage media. Computer-readable media can include, but are not limited to, electronic, magnetic, optical, electromagnetic, infrared or semiconductor systems, devices or devices, or any suitable combination thereof. The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server or data center that includes one or more available media integrations. More detailed examples of computer-readable storage media include electrical connections with one or more wires, magnetic media (e.g., disks, floppy disks, hard disks, tapes, magnetic storage devices), optical media (e.g., optical storage devices, DVDs), semiconductor media (e.g., solid-state hard drives), random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), or any suitable combination thereof, etc.

[0118] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware or any combination thereof. When implemented using software, it can be implemented in whole or in part in the form of a computer program product. The embodiment of the present application also provides at least one computer program product tangibly stored on a non-temporary computer-readable storage medium. The computer program product includes one or more computer executable instructions, such as instructions included in a program module, which are executed in a device on a real or virtual processor of the target to perform the process, method and function involved in any of the above embodiments. When the computer program instruction is loaded and executed on a computer, a process or function according to an embodiment of the present application is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instruction can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium, for example, the computer instruction can be transmitted from a website site, computer, server or data center by wired (e.g., coaxial cable, optical fiber, digital subscriber line) or wireless (e.g., infrared, wireless, microwave, etc.) mode to another website site, computer, server or data center.

[0119] The present application embodiment also proposes a computer program product, including a computer program or instruction, when the computer program or instruction is run on a computer, the computer is made to perform the process, method and function in the above-mentioned embodiment. Usually, a program module includes routines, programs, libraries, objects, classes, components, data structures, etc. that perform specific tasks or realize specific abstract data types. In various embodiments, the functions of program modules can be combined or divided between program modules as needed. Machine executable instructions for program modules can be executed in local or distributed devices. In distributed devices, program modules can be located in local and remote storage media.

[0120] In general, various embodiments of the present application may be implemented in hardware or dedicated circuits, software, logic, or any combination thereof. Some aspects may be implemented in hardware, while other aspects may be implemented in firmware or software, which may be performed by a controller, microprocessor, or other computing device. Although various aspects of the embodiments of the present disclosure are shown and described as block diagrams, flow charts, or using some other graphical representations, it should be understood that the boxes, devices, systems, techniques, or methods described herein may be implemented as, for example, non-limiting examples, hardware, software, firmware, dedicated circuits or logic, general hardware or controllers or other computing devices, or some combination thereof.

[0121] It should be noted that although the embodiments of the present application are described above in conjunction with the accompanying drawings, the above embodiments are not independent of each other, and they can also be combined to obtain other embodiments. The division of the modes, situations, categories and embodiments in the embodiments of the present application is only for the convenience of description and should not constitute a special limitation. The features in the various modes, categories, situations and embodiments can be combined with each other in a logical manner. The various implementation methods of the present application can be combined arbitrarily to achieve different technical effects. The embodiments of the present application no longer list various combinations.

[0122] In addition, although the operation of the method of the present disclosure is described in a particular order in the accompanying drawings, this does not require or imply that these operations must be performed in this particular order, or that all the operations shown must be performed to achieve the desired result. On the contrary, the steps depicted in the flow chart can change the order of execution. Additionally or alternatively, some steps can be omitted, multiple steps can be combined into one step for execution, and / or one step can be decomposed into multiple steps for execution. It should also be noted that the features and functions of two or more devices according to the present disclosure can be embodied in one device. Conversely, the features and functions of a device described above can be further divided into being embodied by multiple devices.

[0123] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

[0124] The above is only an embodiment of the present application and is not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included in the scope of the claims of the present application.

Claims

1. A recommendation method based on user interests, characterized in that: The recommendation method based on user interests includes: Acquire user data, wherein the user data includes attribute data and historical behavior data of the user; Calculating the user's long-term interest score at each point of interest based on the user's historical behavior data and calculating the user's overall interest score based on the user's historical behavior data; According to the long-term interest score of the user at each point of interest and the user data, a multi-objective model based on a user interest efficiency tower is used to predict the short-term interest score of the user at each point of interest; Determining the integrated interest score of the user at each point of interest according to the short-term interest score and the long-term interest score and the overall interest score of the user at each point of interest; Recommend content to users based on their integrated interest scores at various interest points.

2. The method for recommending based on user interests according to claim 1, characterized in that: Calculating the user's long-term interest score at each point of interest based on the user's historical behavior data includes: Counting the user's click data at each point of interest and the exposure data of each point of interest based on the user's historical behavior data; The user's long-term interest score at each point of interest is calculated based on the user's click data at each point of interest and the exposure data at each point of interest.

3. The method for recommending based on user interests according to claim 1, characterized in that: Calculating the overall interest score of the user according to the historical behavior data of the user includes: Counting the user's overall click data and overall exposure data based on the user's historical behavior data; The user's overall interest score is calculated based on the user's overall click data and overall exposure data.

4. The method for recommending based on user interests according to claim 1, characterized in that: Determining the integrated interest score of the user at each interest point according to the short-term interest score, the long-term interest score and the overall interest score of the user at each interest point includes: Compare the short-term interest score with the long-term interest score of each interest point to obtain a first comparison result; Compare the long-term interest score of each interest point with the overall interest score to obtain a second comparison result; The fusion interest score of the user at each point of interest is determined according to the first comparison result and the second comparison result.

5. The method for recommending based on user interests according to claim 4, characterized in that: Determining the fusion interest score of the user at each point of interest according to the first comparison result and the second comparison result includes: If the first comparison result is that the short-term interest score of the interest point is greater than the long-term interest score of the interest point, and the second comparison result is that the long-term interest score of the interest point is greater than the overall interest score, and the difference between the short-term interest score of the interest point and the long-term interest score of the interest point is not less than the difference between the long-term interest score of the interest point and the overall interest score, then the short-term interest score and the long-term interest score of the interest point are merged to obtain a merged interest score of the interest point; If the first comparison result is that the difference between the short-term interest score of the point of interest and the long-term interest score of the point of interest fluctuates within a preset interval, the short-term interest score of the point of interest is used as the fusion interest score of the point of interest, and the preset interval is determined according to the difference between the long-term interest score of the point of interest and the overall interest score; Otherwise, the long-term interest score of the interest point is used as the fusion interest score of the interest point.

6. The method for recommending based on user interests according to claim 1, characterized in that: The recommending content to the user according to the fusion interest points of the user at various interest points includes: Sort the user's integrated interest scores at each point of interest, and recommend content corresponding to the target point of interest to the user based on the sorting result; or, The fused interest score of the user at each point of interest is compared with a preset score threshold to obtain a third comparison result, and content corresponding to the target point of interest is recommended to the user according to the third comparison result.

7. The method for recommending based on user interests according to any one of claims 1 to 6, characterized in that: The multi-objective model based on the user interest efficiency tower includes an expert network, a gated network and an interest efficiency tower. The multi-objective model based on the user interest efficiency tower is trained in the following way: Obtain user sample data and the user's long-term interest score at each point of interest, wherein the user sample data includes the user's attribute data and historical behavior data; Generating user sample feature data according to the user sample data, and inputting the user sample feature data into the expert network to obtain original output features of the expert network; Inputting the user's long-term interest scores at each point of interest into the interest efficiency tower to obtain output features of the interest efficiency tower; Fusing the original output features of the expert network and the output features of the interest efficiency tower to obtain fused features; According to the fusion features, the gating network corresponding to each point of interest is used to predict the user's short-term interest score at each point of interest; The predicted loss value is calculated according to the user's short-term interest score at each point of interest, and the parameters of the multi-objective model based on the user interest efficiency tower are optimized using the predicted loss value to obtain a trained multi-objective model based on the user interest efficiency tower.

8. A recommendation device based on user interests, characterized in that: The recommendation device based on user interests includes: An acquisition unit, used to acquire user data, wherein the user data includes attribute data and historical behavior data of the user; a calculation unit, configured to calculate the user's long-term interest score at each point of interest based on the user's historical behavior data and to calculate the user's overall interest score based on the user's historical behavior data; A prediction unit, configured to predict the user's short-term interest score at each point of interest according to the user's long-term interest score at each point of interest and the user data, using a multi-objective model based on a user interest efficiency tower; A fusion unit, configured to determine a fusion interest score of the user at each interest point according to the short-term interest score and the long-term interest score and the overall interest score of the user at each interest point; The recommendation unit is used to recommend content to users based on the integrated interest scores of users at various interest points.

9. A device comprising: processor; and a memory arranged to store computer executable instructions, wherein when the executable instructions are executed, the processor executes the user interest-based recommendation method according to any one of claims 1 to 7.

10. A computer program product comprising a computer program / instructions, characterized in that When the computer program / instructions are executed by a processor, the recommendation method based on user interests described in any one of claims 1 to 7 is implemented.

Images