An information recommendation method and device, a storage medium, and an electronic device

Abstract

The specification discloses an information recommendation method and device, a storage medium and an electronic device. Based on an information recommendation request sent by a user terminal, candidate commodities are recalled and historical behavior records of the user are determined. Then, for each historical behavior record, an environment association weight corresponding to the historical behavior record is determined according to environment information when the user performs the historical behavior record and current environment information, and an interest feature of the user is determined according to commodity features of commodities corresponding to each historical behavior record and the environment association weight corresponding to the historical behavior record. Finally, a click rate of the user for the candidate commodities is predicted according to the interest feature of the user and the commodity features of the candidate commodities, and information recommendation is performed based on the click rate. By introducing an attention mechanism of environment information, the interest feature of the user is concentrated on the historical behavior record with strong environment association, the interest feature of the user can be more accurately represented, the prediction result is more accurate, and the recommendation hit rate is higher.

Description

Technical Field

[0001] This application relates to the field of information recommendation technology, and in particular to an information recommendation method, apparatus, storage medium and electronic device. Background Technology

[0002] With the development of big data technology, more and more e-commerce platforms are recommending products to users through recommendation systems to improve the user's decision-making experience.

[0003] Currently, when recommending products to users, the click-through rate (CTR) of the recommended products is typically predicted based on the user's historical browsing, clicking, or purchasing records. The products are then displayed and ranked according to the predicted CTR. Specifically, the user's historical behavior records are first obtained, including the products the user has browsed, clicked, or purchased. Next, the product information from these records is input into the pooling layer of a click prediction model for average pooling, and the pooling result is used as the user's interest feature. Then, this interest feature, along with the product information of the recommended products, is input into the fully connected layer of the click prediction model to predict the user's CTR for those products.

[0004] However, in location-based services (LBS) scenarios, such as food delivery, user behavior records may be distributed across multiple geographical locations. The products retrieved by the recommendation system are limited by LBS, meaning they can only retrieve products near the user's current location. In this case, predicting the click-through rate of the retrieved products based on the user's behavior records across different geographical locations often results in low accuracy, leading to a low hit rate for information recommendations. Summary of the Invention

[0005] This specification provides an information recommendation method, apparatus, storage medium, and electronic device to partially address problems in the prior art.

[0006] The embodiments in this specification adopt the following technical solutions:

[0007] This specification provides an information recommendation method, including:

[0008] Receive an information recommendation request sent by a user terminal, wherein the information recommendation request carries the user's user identifier and current environment information;

[0009] Based on the current environmental information, recall the candidate products;

[0010] Based on the user identifier, the user's historical behavior records and the environmental information in which the user performed each historical behavior record are obtained, wherein the environmental information includes at least one of geographical location and time;

[0011] For each historical behavior record of the user, the environmental information when the user performed the historical behavior record and the user's current environmental information are input into the first activation unit layer of the pre-trained click prediction model to determine the environmental association weight corresponding to the historical behavior record.

[0012] The user's interest characteristics are determined based on the product characteristics of the products corresponding to each of the user's historical behavior records and the environmental association weights corresponding to each of the historical behavior records.

[0013] The user's interest characteristics and the product characteristics of the candidate products are input into the fully connected layer of the click prediction model to predict the user's click-through rate on the candidate products, so as to make information recommendations based on the click-through rate of the candidate products.

[0014] Optionally, before determining the user's interest characteristics, the method further includes:

[0015] The product features of the candidate products and the product features of the products corresponding to the historical behavior record are input into the second activation unit layer of the click prediction model to determine the product association weight corresponding to the historical behavior record.

[0016] The product association weight represents the degree of association between the product corresponding to the historical behavior record and the candidate product;

[0017] Based on the product characteristics of the corresponding items in each of the user's historical behavior records, and the environmental association weights corresponding to each historical behavior record, the user's interest characteristics are determined, specifically including:

[0018] The user's interest characteristics are determined based on the product characteristics of the products corresponding to each of the user's historical behavior records, the environmental association weights corresponding to each of the historical behavior records, and the product association weights corresponding to each of the historical behavior records.

[0019] Optionally, the environmental information of the user when performing the historical behavior record and the user's current environmental information are input into the first activation unit layer of the pre-trained click prediction model to determine the environmental association weights corresponding to the historical behavior record, specifically including:

[0020] Based on the environmental information when the user executed the historical behavior record and the user's current environmental information, determine the environmental association features;

[0021] The environmental information when the user performed the historical behavior record, the user's current environmental information, and the environmental association features are input into the feedforward neural network layer of the first activation unit layer in the pre-trained click prediction model to determine the environmental association weights corresponding to the historical behavior record.

[0022] The environmental association weight represents the degree of association between the environmental information when the user performed the historical behavior record and the current environmental information.

[0023] Optionally, the product features of the candidate product and the product features of the product corresponding to the historical behavior record are input into the second activation unit layer of the click prediction model to determine the product association weight corresponding to the historical behavior record, specifically including:

[0024] Based on the product characteristics of the candidate products and the product characteristics of the products corresponding to the historical behavior records, product association characteristics are determined;

[0025] The product features of the candidate product, the product features of the product corresponding to the historical behavior record, and the product association features are input into the feedforward neural network layer of the second activation unit layer in the click prediction model to determine the product association weight corresponding to the historical behavior record.

[0026] Optionally, the user's historical behavior records include at least one of the following: historical browsing records of products, historical clicking records of products, historical purchasing records of products, and historical favorite records of products.

[0027] Optionally, the click prediction model further includes an encoding layer;

[0028] The method further includes:

[0029] The product information of the candidate products is input into the encoding layer of the click prediction model to determine the product characteristics of the candidate products;

[0030] The product information of the products corresponding to each of the user's historical behavior records is input into the encoding layer of the click prediction model to determine the product features of the products corresponding to each historical behavior record;

[0031] The environmental information of the user when performing the historical behavior record and the user's current environmental information are input into the first activation unit layer of the pre-trained click prediction model, specifically including:

[0032] The environmental information of the user when performing the historical behavior record is input into the encoding layer of the click prediction model to determine the environmental characteristics of the user when performing the historical behavior record;

[0033] The user's current environment information is input into the encoding layer of the click prediction model to determine the user's current environment characteristics;

[0034] The environmental characteristics of the user when the user performed the historical behavior record and the current environmental characteristics of the user are input into the first activation unit layer of the click prediction model.

[0035] Optionally, training a click prediction model includes:

[0036] The training sample set is determined by identifying the products that were recommended and displayed to each user in history, the environmental information when the products were recommended and displayed, the historical behavior records of each user, and the environmental information when each user executed the historical behavior records. For each training sample in the training sample set, the training sample is labeled according to the click information of the user on the recommended and displayed products in the training sample.

[0037] For each historical behavior record of a user in the training sample, the environmental information when the user performed the historical behavior record and the environmental information when the recommendation was shown to the user are input into the first activation unit layer of the click prediction model to be trained to determine the prediction environment association weight corresponding to the historical behavior record.

[0038] The user's interest characteristics are determined based on the product features of the products corresponding to each of the user's historical behavior records and the prediction environment association weights corresponding to each of the historical behavior records.

[0039] The user's interest features and the product features of the products recommended to the user are input into the fully connected layer of the click prediction model to be trained to determine the user's predicted click-through rate for the recommended products.

[0040] With the goal of minimizing the difference between the predicted click-through rate of users for recommended products in each training sample and the labeling of each training sample, the model parameters of each network layer in the click prediction model are adjusted.

[0041] This specification provides an information recommendation device, comprising:

[0042] The receiving module is configured to receive an information recommendation request sent by a user terminal, wherein the information recommendation request carries the user's user identifier and current environment information.

[0043] The recall module is configured to recall candidate products based on the current environmental information.

[0044] The acquisition module is configured to acquire the user's historical behavior records and the environmental information in which the user performed each historical behavior record based on the user identifier. The environmental information includes at least one of geographical location and time.

[0045] The activation module is configured to input the environmental information of the user when the user performed the historical behavior record and the user's current environmental information into the first activation unit layer of the pre-trained click prediction model for each historical behavior record, and determine the environmental association weight corresponding to the historical behavior record.

[0046] The determination module is configured to determine the user's interest characteristics based on the product characteristics of the products corresponding to each of the user's historical behavior records and the environmental association weights corresponding to each of the historical behavior records.

[0047] The prediction module is configured to input the user's interest features and the product features of the candidate products into the fully connected layer of the click prediction model to predict the user's click-through rate on the candidate products, so as to make information recommendations based on the click-through rate of the candidate products.

[0048] This specification provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the aforementioned information recommendation method.

[0049] This specification provides an electronic device including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the aforementioned information recommendation method.

[0050] The above-described at least one technical solution adopted in the embodiments of this specification can achieve the following beneficial effects:

[0051] This specification describes a method for recalling candidate products and identifying user historical behavior records based on information recommendation requests sent from user terminals. For each historical behavior record, the environmental information at the time the user executed that behavior, along with the current environmental information, is used to determine the environmental association weight corresponding to that historical behavior record through the first activation unit layer of the click prediction model. Furthermore, the user's interest characteristics are determined based on the product features of each historical behavior record and the corresponding environmental association weights. Finally, based on the user's interest characteristics and the product features of the candidate products, the click-through rate of the user for the candidate products is predicted through the fully connected layer of the click prediction model, enabling information recommendation based on the click-through rate. By introducing an attention mechanism for environmental information, the user's interest characteristics are concentrated on historical behavior records with strong environmental associations, resulting in a more accurate representation of the user's interest characteristics, more accurate prediction results, and a higher recommendation hit rate. Attached Figure Description

[0052] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings:

[0053] Figure 1 A flowchart illustrating an information recommendation method provided in an embodiment of this specification;

[0054] Figure 2This is a schematic diagram of the network structure of the first activation unit layer provided in the embodiments of this specification;

[0055] Figure 3 A network structure diagram of a click prediction model provided in the embodiments of this specification;

[0056] Figure 4 A network structure diagram of a click prediction model provided in the embodiments of this specification;

[0057] Figure 5 This is a schematic diagram of the structure of an information recommendation device provided in an embodiment of this specification;

[0058] Figure 6 A schematic diagram of an electronic device for implementing the information recommendation method provided in the embodiments of this specification. Detailed Implementation

[0059] To make the objectives, technical solutions, and advantages of this specification clearer, the technical solutions of this application will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments in this specification without creative effort are within the scope of protection of this application.

[0060] Currently, when e-commerce platforms recommend products to users through their recommendation systems, they can predict the click-through rate of each product based on the user's historical browsing information and the product information of the product to be recommended. Based on this click-through rate, product recommendations are then made. Since e-commerce platforms provide non-instant delivery services, using methods such as express delivery, the recalled products for recommendation are often not limited by geographical location, time, or other factors.

[0061] However, in location-based service scenarios, taking food delivery as an example, a user's browsing history includes products viewed at home and at the office. If a user orders food delivery at the office, the delivery platform's recommendation system, limited by their current geographical location, can only recall a few candidate products near the office. In this case, predicting the click-through rate of each recalled candidate product based on all products the user has browsed at home and at the office is often inaccurate.

[0062] Furthermore, in the food delivery scenario, users' needs differ across morning, noon, and evening. A user's browsing history includes breakfast, lunch, and dinner viewed at different times. If a user places an order at noon, the click-through rate prediction for the currently recalled candidate products based on the user's browsing history at different times will also show significant differences.

[0063] In view of the above-mentioned technical problems, this specification provides an information recommendation method. The technical solutions provided by the various embodiments of this application are described in detail below with reference to the accompanying drawings.

[0064] Figure 1 This is a flowchart illustrating an information recommendation method provided in an embodiment of this specification, which may specifically include the following steps:

[0065] S100: Receives information recommendation requests sent by user terminals.

[0066] S102: Based on the current environmental information, recall the candidate products.

[0067] S104: Based on the user identifier, obtain the user's historical behavior records and the environmental information in which the user performed each historical behavior record.

[0068] The information recommendation method provided in this manual can be applied to application scenarios such as search ranking or homepage recommendations. It recommends products to users by predicting the click-through rate of each candidate product.

[0069] The information recommendation method can be executed by the server of the recommendation system. The server can be a single server or a system composed of multiple servers, such as a distributed server system. This specification does not impose any restrictions on this and can be configured as needed.

[0070] Specifically, the server can receive information recommendation requests sent by user terminals. These requests can be either homepage loading requests or search requests. When the request is a homepage loading request, it carries a user identifier and the user's current environment information; the user identifier can be the user's account identifier. When the request is a search request, it also carries search keywords.

[0071] The server can then recall several candidate products based on the user's current environmental information in the recommendation request. The current environmental information includes at least one of the user's current geographical location and current time. For example, it can recall nearby merchants within a 2km radius of the current geographical location and use their products as candidate products. Alternatively, it can recall merchants operating during the current time period and use their products as candidate products.

[0072] Furthermore, if the information recommendation request is a search request, then it is also necessary to filter products that match the keywords searched by the user from among the candidate products.

[0073] Furthermore, when predicting the click-through rate of users for each candidate product, predictions can be made based on user interests, which are related to the user's historical behavior. Therefore, it is also necessary to retrieve the user's historical behavior records and the environmental information in which the user performed each historical behavior record from the stored historical behavior data, based on the user identifier in the information recommendation request. The user's historical behavior records include at least one of the following: historical product browsing records, historical product clicking records, historical product purchase records, and historical product favorites records.

[0074] Furthermore, since users' interests and preferences change over time, in order to obtain more accurate prediction results, we can obtain the user's historical behavior data over a recent period, such as the most recent week or the most recent month.

[0075] It should be noted that there is no restriction on the execution order of the above steps S102 and S104, and they can be executed simultaneously.

[0076] S106: For each historical behavior record of the user, input the environmental information when the user performed the historical behavior record and the user's current environmental information into the first activation unit layer of the pre-trained click prediction model to determine the environmental association weight corresponding to the historical behavior record.

[0077] User historical behavior data includes user behavior in various environments. For example, based on different geographical locations, there are products that users browse at home and products that users browse at work; based on different time periods, there are products that users browse at noon and products that users browse in the evening; and based on different weekdays and products that users browse on weekends.

[0078] When predicting click-through rates for candidate products, it's also necessary to consider the impact of varying environmental information on user interests and preferences. For example, if a user is currently ordering takeout at work, the products they previously viewed at work have a greater weight in influencing whether they click on the candidate product. Conversely, if a user is currently ordering on a weekend, the products they previously viewed on weekends have a greater weight in influencing whether they click on the candidate product.

[0079] Therefore, when determining user interests based on historical behavior, it is also necessary to consider the impact of environmental information on user preferences. For example, users may prefer fast food for lunch and noodles for dinner.

[0080] Specifically, for each historical behavior record of a user, the environmental information in which the user performed that historical behavior can be determined, such as the geographical location or time period when the user browsed product A in the past. Then, the environmental information of the user when performing that historical behavior record and the user's current environmental information are input into the first activation unit layer of the pre-trained click prediction model to determine the environmental association weights corresponding to that historical behavior record.

[0081] The click prediction model also includes an encoding layer. Before inputting into the first activation unit layer, the environmental information of the user when executing the historical behavior record and the user's current environmental information can be input into the encoding layer to obtain the environmental characteristics of the user when executing the historical behavior record and the user's current environmental characteristics.

[0082] Furthermore, the network structure of the first activation unit layer is as follows: Figure 2 As shown, after inputting the environmental information of the user when executing the historical behavior record and the user's current environmental information into the first activation unit layer, environmental association features can be determined based on these two information. Then, the environmental information of the user when executing the historical behavior record, the user's current environmental information, and the environmental association features are input into the feedforward neural network (FNN) layer of the first activation unit layer to determine the environmental association weights corresponding to the historical behavior record. These environmental association weights characterize the degree of association between the environmental information of the user when executing the historical behavior record and the current environmental information.

[0083] S108: Determine the user's interest characteristics based on the product characteristics of the products corresponding to each of the user's historical behavior records and the environmental association weights corresponding to each of the historical behavior records.

[0084] After determining the environmental association weights between the environmental information of each historical behavior record and the current environmental information, an attention mechanism can be used to filter and reduce the weight of each user's historical behavior record, and extract feature information that is more suitable for LBS scenarios.

[0085] Specifically, for each historical behavior record of a user, the product characteristics of the corresponding product can be determined. Then, based on the product characteristics of the corresponding product and the environmental association weight of the corresponding historical behavior record, the user's interest characteristics are determined.

[0086] In determining the product characteristics of each historical behavior record, the product information of each product corresponding to each historical behavior record can be obtained first. This product information is then input into the encoding layer of the click prediction model to obtain the product characteristics of each product. The product characteristics include at least one of the following: the product identifier, the merchant identifier, and the category identifier.

[0087] Furthermore, since a user's interest characteristics are not only related to historical behavior records, the user's interest characteristics can also be determined by combining the user's profile information and historical behavior records.

[0088] S110: Input the user's interest features and the product features of the candidate products into the fully connected layer of the click prediction model to predict the user's click rate on the candidate products, so as to make information recommendations based on the click rate on the candidate products.

[0089] After extracting user interest features based on historical user behavior records, the click-through rate of a candidate product can be predicted based on the user's interest features and the product features of the candidate product, so as to make information recommendations.

[0090] Specifically, for each candidate product, the user's interest features are concatenated with the product features of that candidate product, and then input into the fully connected layer of the click prediction model to predict the user's click-through rate for that candidate product. Then, based on the user's ranking of the click-through rates for each candidate product, the recommended information to be displayed to the user is determined.

[0091] Figure 3 The network structure diagram of a click prediction model provided in this embodiment allows the following steps: The model inputs the product information of candidate products to be recommended, the product information of products corresponding to the user's historical behavior records, the environmental information when the user executed historical behavior records, and the current environmental information into the encoding layer of the click prediction model. This yields the product features of the candidate products, the product features of products corresponding to the user's historical behavior records, the environmental features when the user executed historical behavior records, and the current environmental features. Then, the environmental features when the user executed historical behavior records and the current environmental features are input into the first activation unit layer of the click prediction model to obtain the environmental association weights output by the first activation unit layer. Next, based on the product features of the products corresponding to the user's historical behavior records and the environmental association weights, the user's interest features are determined. Finally, the user's interest features and the product features of the candidate products are input into the fully connected layer of the click prediction model to obtain the user's click-through rate for the candidate products.

[0092] based on Figure 1The information recommendation method shown can recall candidate products and determine the user's historical behavior records based on information recommendation requests sent by the user terminal. Then, for each historical behavior record, based on the environmental information when the user performed the historical behavior and the current environmental information, the environmental association weight corresponding to the historical behavior record is determined through the first activation unit layer of the click prediction model. Furthermore, based on the product features of the products corresponding to each historical behavior record and the environmental association weights, the user's interest features are determined. Finally, based on the user's interest features and the product features of the candidate products, the click-through rate of the user for the candidate products is predicted through the fully connected layer of the click prediction model, enabling information recommendation based on the click-through rate. By introducing an attention mechanism for environmental information, the user's interest features are concentrated on historical behavior records with strong environmental associations, which can more accurately represent the user's interest features, resulting in more accurate prediction results and a higher recommendation hit rate.

[0093] This manual describes how to train the click prediction model described above, which may include the following steps:

[0094] S200: Determine the products that were recommended and displayed to each user in history, the environmental information when the products were recommended and displayed, the historical behavior records of each user before the products were recommended and displayed, and the environmental information when each user executed the historical behavior records, as a training sample set. For each training sample in the training sample set, label the training sample according to the click information of the users on the recommended and displayed products in the training sample.

[0095] First, for each user, we can obtain the products historically recommended to that user, as well as the environmental information when those recommendations were displayed. We also obtain the user's historical behavior records before the recommendations were displayed, and the environmental information when those historical behaviors were executed, as training samples. Then, we label the samples based on the user's click history on the recommended products.

[0096] S202: For each historical behavior record of a user in the training sample, input the environmental information when the user performed the historical behavior record and the environmental information when the recommendation was displayed to the user into the first activation unit layer of the click prediction model to be trained, and determine the prediction environment association weight corresponding to the historical behavior record.

[0097] The click prediction model to be trained also includes an encoding layer, which encodes environmental and product information into corresponding vector features according to preset encoding rules. Therefore, before inputting into the first activation unit layer, the encoding layer is needed to encode the environmental information when the user performed the historical behavior record, as well as the current environmental information, into environmental features.

[0098] S204: Determine the user's interest features based on the product characteristics of the products corresponding to each of the user's historical behavior records and the prediction environment association weights corresponding to each of the historical behavior records.

[0099] S206: Input the user's interest features and the product features of the products recommended to the user into the fully connected layer of the click prediction model to be trained, and determine the predicted click-through rate of the user for the recommended products.

[0100] S208: With the goal of minimizing the difference between the predicted click-through rate of users for the recommended products in each training sample and the labeling of each training sample, adjust the model parameters of each network layer in the click prediction model.

[0101] Finally, based on the click-through rate of users for the recommended products output by the click prediction model and the annotation of each training sample, the model parameters of each network layer in the click prediction model can be adjusted with the goal of minimizing the difference between the two.

[0102] In one embodiment of this specification, it is assumed that the product characteristics of the products corresponding to each historical behavior record of user U are {e1, e2, e3...e...} H The environmental characteristics of user U when executing each historical behavior record are {C1, C2, C3...C}. H}, where e1, e2…e H These correspond to the products associated with each historical behavior record, C1, C2…C H Each corresponds to a historical behavior record, U C This indicates the current environmental characteristics of the user.

[0103] Then, the environmental association weight of each historical behavior record can be obtained through the first activation unit layer as f1(U C C j (j) represents any record in the historical behavior log. Based on the product characteristics of the goods corresponding to each historical behavior record and the environmental association weights corresponding to each historical behavior record, the user's interest characteristics can be determined.

[0104] Furthermore, due to the diversity of user interests, not all products a user browses influence their current click target. For example, if the current target is a mouse, and the user's past browsing includes keyboards, clothes, and water bottles, only the keyboard will influence whether the user clicks on the mouse. In other words, the weight of influence from the user's past browsing on their current click target varies.

[0105] Therefore, in another embodiment of this specification, a product association attention mechanism can be introduced to focus the user's interest characteristics on historical behavior records with high product association.

[0106] Specifically, the click prediction model also includes a second activation unit layer, used to determine the correlation between each product corresponding to a historical behavior record and the candidate product. The product features of the candidate product and the product features of the product corresponding to the historical behavior record are input into the second activation unit layer of the click prediction model to determine the product association weight corresponding to the historical behavior record. This product association weight represents the degree of correlation between the product corresponding to the historical behavior record and the candidate product.

[0107] Then, based on the product characteristics of the products corresponding to each of the user's historical behavior records, the environmental association weights and product association weights corresponding to each of the historical behavior records, the user's interest characteristics are determined.

[0108] Continuing with the example above, let's assume that the product characteristic of candidate product A is V. A Then, the environmental association weight of each historical behavior record can be obtained through the second activation unit layer as f2(V A e j (j) represents any record in the historical behavior log. Based on the product characteristics of the goods corresponding to each historical behavior record, the environmental association weights of each historical behavior record, and the product association weights, the user's interest characteristics can be determined.

[0109] Figure 4 The network structure diagram of a click prediction model provided in the embodiments of this specification can be input into the encoding layer of the click prediction model with the product information of the candidate product to be recommended, the product information of the product corresponding to the user's historical behavior record, the environmental information when the user executes the historical behavior record, and the current environmental information, so as to obtain the product features of the candidate product, the product features of the product corresponding to the user's historical behavior record, the environmental features when the user executes the historical behavior record, and the current environmental features.

[0110] Next, the environmental features of the user's historical behavior records and the current environmental features are input into the first activation unit layer of the click prediction model to obtain the environmental association weights output by the first activation unit layer. Then, the product features of the candidate products and the product features corresponding to the historical behavior records are input into the second activation unit layer to obtain the product association weights output by the second activation unit layer. Then, based on the product features of the products corresponding to the user's historical behavior records, the environmental association weights, and the product association weights, the user's interest features are determined. Finally, the user's interest features and the product features of the candidate products are input into the fully connected layer of the click prediction model to obtain the user's click-through rate for the candidate products.

[0111] In this second activation unit layer, product association features are first determined based on the product features of the candidate product and the product features of the product corresponding to the historical behavior record. Then, the product features of the candidate product, the product features of the product corresponding to the historical behavior record, and the product association features are input into the FNN layer of the second activation unit layer to obtain the product association weights corresponding to the historical behavior record.

[0112] In this specification, a method for determining the environmental association weight of each historical behavior record based on the environmental information when the user executes each historical behavior record and the current environmental information is used. Compared with the method of rigidly isolating historical behavior records according to environmental conditions, this method can uncover some potential interests of users in different scenarios, making the user interest characteristics richer.

[0113] based on Figure 1 The present invention provides a schematic diagram of an information recommendation device, as illustrated in the embodiments of this specification. Figure 5 As shown.

[0114] Figure 5 A schematic diagram of an information recommendation device provided in the embodiments of this specification includes:

[0115] The receiving module 300 is configured to receive an information recommendation request sent by a user terminal, wherein the information recommendation request carries the user's user identifier and current environment information.

[0116] Recall module 302 is configured to recall candidate products based on the current environmental information;

[0117] The acquisition module 304 is configured to acquire the user's historical behavior records and the environmental information in which the user performed each historical behavior record based on the user identifier. The environmental information includes at least one of geographical location and time.

[0118] Activation module 306 is configured to input the environmental information of the user when the user performed the historical behavior record and the user's current environmental information into the first activation unit layer of the pre-trained click prediction model for each historical behavior record, and determine the environmental association weight corresponding to the historical behavior record.

[0119] The determination module 308 is configured to determine the user's interest characteristics based on the product characteristics of the products corresponding to each of the user's historical behavior records and the environmental association weights corresponding to each of the historical behavior records.

[0120] The prediction module 310 is configured to input the user's interest features and the product features of the candidate products into the fully connected layer of the click prediction model to predict the user's click rate on the candidate products, so as to make information recommendations based on the click rate on the candidate products.

[0121] Optionally, the activation module 306 is further configured to input the product features of the candidate product and the product features of the product corresponding to the historical behavior record into the second activation unit layer of the click prediction model to determine the product association weight corresponding to the historical behavior record, wherein the product association weight represents the degree of association between the product corresponding to the historical behavior record and the candidate product. The determination module 308 is specifically configured to determine the user's interest features based on the product features of the products corresponding to each of the user's historical behavior records, the environmental association weight corresponding to each of the historical behavior records, and the product association weight corresponding to each of the historical behavior records.

[0122] Optionally, the activation module 306 is specifically used to determine environmental association features based on the environmental information when the user performed the historical behavior record and the user's current environmental information, and input the environmental information when the user performed the historical behavior record, the user's current environmental information, and the environmental association features into the feedforward neural network layer of the first activation unit layer in the pre-trained click prediction model to determine the environmental association weight corresponding to the historical behavior record, wherein the environmental association weight represents the degree of association between the environmental information when the user performed the historical behavior record and the current environmental information.

[0123] Optionally, the activation module 306 is further configured to determine product association features based on the product features of the candidate product and the product features of the product corresponding to the historical behavior record, and input the product features of the candidate product, the product features of the product corresponding to the historical behavior record, and the product association features into the feedforward neural network layer of the second activation unit layer in the click prediction model to determine the product association weight corresponding to the historical behavior record.

[0124] Optionally, the user's historical behavior records include at least one of the following: historical browsing records of products, historical clicking records of products, historical purchasing records of products, and historical favorite records of products.

[0125] Optionally, the activation module 306 is further configured to input the product information of the candidate product into the encoding layer of the click prediction model to determine the product features of the candidate product, input the product information of the product corresponding to each historical behavior record of the user into the encoding layer of the click prediction model to determine the product features of the product corresponding to each historical behavior record, and specifically, the activation module 306 is configured to input the environmental information of the user when executing the historical behavior record into the encoding layer of the click prediction model to determine the environmental features of the user when executing the historical behavior record, input the current environmental information of the user into the encoding layer of the click prediction model to determine the current environmental features of the user, and input the environmental features of the user when executing the historical behavior record and the current environmental features of the user into the first activation unit layer of the click prediction model.

[0126] Optionally, the information recommendation device further includes a model training module 312. Specifically, the model training module 312 is used to determine the products historically recommended to each user, the environmental information at the time of recommendation, the user's historical behavior records before the recommendation, and the environmental information when each user executed historical behavior records, as a training sample set. For each training sample in the training sample set, the module labels the training sample based on the user's clicks on the recommended products in that training sample. For each historical behavior record of a user in the training sample, the module inputs the environmental information when the user executed that historical behavior record and the environmental information when the product was recommended to the user into the training module. The first activation unit layer of the click prediction model determines the prediction environment association weights corresponding to the historical behavior record. Based on the product features of the products corresponding to each historical behavior record of the user, and the prediction environment association weights corresponding to each historical behavior record, the user's interest features are determined. The user's interest features and the product features of the products recommended to the user are input into the fully connected layer of the click prediction model to be trained to determine the predicted click-through rate of the user for the recommended products. With the goal of minimizing the difference between the predicted click-through rate of the user for the recommended products in each training sample and the labeling of each training sample, the model parameters of each network layer in the click prediction model are adjusted.

[0127] This specification also provides a computer-readable storage medium storing a computer program that can be used to execute the above-described embodiments. Figure 1 Information recommendation methods are provided.

[0128] according to Figure 1 The information recommendation method shown in this specification also includes embodiments that propose... Figure 6 The diagram shows a schematic structural representation of the electronic device. Figure 6At the hardware level, this electronic device includes a processor, internal bus, network interface, memory, and non-volatile memory, and may also include other hardware required for business operations. The processor reads the corresponding computer program from the non-volatile memory into memory and then executes it to achieve the above. Figure 1 The information recommendation method shown.

[0129] Of course, in addition to software implementation, this specification does not exclude other implementation methods, such as logic devices or a combination of hardware and software. In other words, the execution subject of the following processing flow is not limited to each logic unit, but can also be hardware or logic devices.

[0130] In the 1990s, improvements to a technology could be clearly distinguished as either hardware improvements (e.g., improvements to the circuit structure of diodes, transistors, switches, etc.) or software improvements (improvements to the methodology). However, with technological advancements, many methodological improvements today can be considered direct improvements to the hardware circuit structure. Designers almost always obtain the corresponding hardware circuit structure by programming the improved methodology into the hardware circuit. Therefore, it cannot be said that a methodological improvement cannot be implemented using hardware physical modules. For example, a Programmable Logic Device (PLD) (e.g., a Field Programmable Gate Array (FPGA)) is such an integrated circuit whose logic function is determined by the user programming the device. Designers can program and "integrate" a digital system onto a PLD themselves, without needing chip manufacturers to design and manufacture dedicated integrated circuit chips. Moreover, nowadays, instead of manually generating integrated circuit chips, this programming is mostly implemented using "logic compiler" software. Similar to the software compiler used in program development, the original code before compilation must be written in a specific programming language, called a Hardware Description Language (HDL). There are many HDLs, such as ABEL (Advanced Boolean Expression Language), AHDL (Altera Hardware Description Language), Confluence, CUPL (Cornell University Programming Language), HDCal, JHDL (Java Hardware Description Language), Lava, Lola, MyHDL, PALASM, and RHDL (Ruby Hardware Description Language). Currently, the most commonly used are VHDL (Very-High-Speed ​​Integrated Circuit Hardware Description Language) and Verilog. Those skilled in the art should also understand that by simply performing some logic programming on the method flow using one of these hardware description languages ​​and programming it into an integrated circuit, the hardware circuit implementing the logical method flow can be easily obtained.

[0131] The controller can be implemented in any suitable manner. For example, it can take the form of a microprocessor or processor and a computer-readable medium storing computer-readable program code (e.g., software or firmware) executable by the (micro)processor, logic gates, switches, application-specific integrated circuits (ASICs), programmable logic controllers, and embedded microcontrollers. Examples of controllers include, but are not limited to, the following microcontrollers: ARC 625D, Atmel AT91SAM, Microchip PIC18F26K20, and Silicon Labs C8051F320. A memory controller can also be implemented as part of the control logic of the memory. Those skilled in the art will also recognize that, in addition to implementing the controller in purely computer-readable program code form, the same functionality can be achieved by logically programming the method steps to make the controller take the form of logic gates, switches, application-specific integrated circuits, programmable logic controllers, and embedded microcontrollers. Therefore, such a controller can be considered a hardware component, and the means included therein for implementing various functions can also be considered as structures within the hardware component. Alternatively, the means for implementing various functions can be considered as both software modules implementing the method and structures within the hardware component.

[0132] The systems, devices, modules, or units described in the above embodiments can be implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer. Specifically, a computer can be, for example, a personal computer, laptop computer, cellular phone, camera phone, smartphone, personal digital assistant, media player, navigation device, email device, game console, tablet computer, wearable device, or any combination of these devices.

[0133] For ease of description, the above devices are described in terms of function, divided into various units. Of course, in implementing this specification, the functions of each unit can be implemented in one or more software and / or hardware components.

[0134] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0135] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0136] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0137] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0138] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0139] Memory may include non-persistent storage in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0140] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0141] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0142] Those skilled in the art will understand that the embodiments of this specification can be provided as methods, systems, or computer program products. Therefore, this specification may take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this specification may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0143] This specification can be described in the general context of computer-executable instructions that are executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc., that perform a specific task or implement a specific abstract data type. This specification can also be practiced in distributed computing environments, where tasks are performed by remote processing devices connected via a communication network. In distributed computing environments, program modules can reside in local and remote computer storage media, including storage devices.

[0144] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to interchangeably. Each embodiment focuses on describing the differences from other embodiments. In particular, the system embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions in the method embodiments.

[0145] The above description is merely an embodiment of this specification and is not intended to limit this specification. Various modifications and variations can be made to this specification by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this specification should be included within the scope of the claims of this specification.

Claims

1. An information recommendation method, characterized in that, include: Receive an information recommendation request sent by a user terminal, wherein the information recommendation request carries the user's user identifier and current environment information; Based on the current environmental information, recall the candidate products; Based on the user identifier, the user's historical behavior records and the environmental information in which the user performed each historical behavior record are obtained, wherein the environmental information includes at least one of geographical location and time; For each historical behavior record of the user, the environmental information when the user performed the historical behavior record and the user's current environmental information are input into the first activation unit layer of the pre-trained click prediction model to determine the environmental association weight corresponding to the historical behavior record. The product features of the candidate products and the product features of the products corresponding to the historical behavior record are input into the second activation unit layer of the click prediction model to determine the product association weight corresponding to the historical behavior record. The product association weight represents the degree of association between the product corresponding to the historical behavior record and the candidate product; The user's interest characteristics are determined based on the product characteristics of the products corresponding to each of the user's historical behavior records, the environmental association weights corresponding to each of the historical behavior records, and the product association weights corresponding to each of the historical behavior records. The user's interest characteristics and the product characteristics of the candidate products are input into the fully connected layer of the click prediction model to predict the user's click-through rate on the candidate products, so as to make information recommendations based on the click-through rate of the candidate products.

2. The method according to claim 1, characterized in that, The environmental information of the user when performing the historical behavior record and the user's current environmental information are input into the first activation unit layer of the pre-trained click prediction model to determine the environmental association weights corresponding to the historical behavior record, specifically including: Based on the environmental information when the user executed the historical behavior record and the user's current environmental information, determine the environmental association features; The environmental information when the user performed the historical behavior record, the user's current environmental information, and the environmental association features are input into the feedforward neural network layer of the first activation unit layer in the pre-trained click prediction model to determine the environmental association weights corresponding to the historical behavior record. The environmental association weight represents the degree of association between the environmental information when the user performed the historical behavior record and the current environmental information.

3. The method according to claim 1, characterized in that, The product features of the candidate products and the product features of the products corresponding to the historical behavior record are input into the second activation unit layer of the click prediction model to determine the product association weight corresponding to the historical behavior record, specifically including: Based on the product characteristics of the candidate products and the product characteristics of the products corresponding to the historical behavior records, product association characteristics are determined; The product features of the candidate product, the product features of the product corresponding to the historical behavior record, and the product association features are input into the feedforward neural network layer of the second activation unit layer in the click prediction model to determine the product association weight corresponding to the historical behavior record.

4. The method according to claim 1, characterized in that, The user's historical behavior records include at least one of the following: historical browsing records, historical clicking records, historical purchasing records, and historical favorited records.

5. The method according to claim 1, characterized in that, The click prediction model also includes an encoding layer; The method further includes: The product information of the candidate products is input into the encoding layer of the click prediction model to determine the product characteristics of the candidate products; The product information of the products corresponding to each of the user's historical behavior records is input into the encoding layer of the click prediction model to determine the product characteristics of the products corresponding to each historical behavior record; The environmental information of the user when performing the historical behavior record and the user's current environmental information are input into the first activation unit layer of the pre-trained click prediction model, specifically including: The environmental information of the user when performing the historical behavior record is input into the encoding layer of the click prediction model to determine the environmental characteristics of the user when performing the historical behavior record; The user's current environment information is input into the encoding layer of the click prediction model to determine the user's current environment characteristics; The environmental characteristics of the user when the user performed the historical behavior record and the current environmental characteristics of the user are input into the first activation unit layer of the click prediction model.

6. The method according to claim 1, characterized in that, Training the click prediction model specifically includes: The training sample set is determined by identifying the products that were recommended and displayed to each user in history, the environmental information at the time of the recommendation and display, the historical behavior records of each user before the recommendation and display, and the environmental information when each user executed the historical behavior records. For each training sample in the training sample set, the training sample is labeled according to the click information of the user on the recommended and displayed products in the training sample. For each historical behavior record of a user in the training sample, the environmental information when the user performed the historical behavior record and the environmental information when the recommendation was shown to the user are input into the first activation unit layer of the click prediction model to be trained to determine the prediction environment association weight corresponding to the historical behavior record. The user's interest characteristics are determined based on the product features of the products corresponding to each of the user's historical behavior records and the prediction environment association weights corresponding to each of the historical behavior records. The user's interest features and the product features of the products recommended to the user are input into the fully connected layer of the click prediction model to be trained to determine the user's predicted click-through rate for the recommended products. With the goal of minimizing the difference between the predicted click-through rate of users for recommended products in each training sample and the labeling of each training sample, the model parameters of each network layer in the click prediction model are adjusted.

7. An information recommendation device, characterized in that, include: The receiving module is configured to receive an information recommendation request sent by a user terminal, wherein the information recommendation request carries the user's user identifier and current environment information. The recall module is configured to recall candidate products based on the current environmental information. The acquisition module is configured to acquire the user's historical behavior records and the environmental information in which the user performed each historical behavior record based on the user identifier. The environmental information includes at least one of geographical location and time. The activation module is configured to, for each historical behavior record of the user, input the environmental information of the user when the historical behavior record was executed and the current environmental information of the user into the first activation unit layer of the pre-trained click prediction model to determine the environmental association weight corresponding to the historical behavior record; and input the product features of the candidate product and the product features of the product corresponding to the historical behavior record into the second activation unit layer of the click prediction model to determine the product association weight corresponding to the historical behavior record. The product association weight represents the degree of association between the product corresponding to the historical behavior record and the candidate product; The determination module is configured to determine the user's interest characteristics based on the product characteristics of the products corresponding to each of the user's historical behavior records, the environmental association weights corresponding to each of the historical behavior records, and the product association weights corresponding to each of the historical behavior records. The prediction module is configured to input the user's interest features and the product features of the candidate products into the fully connected layer of the click prediction model to predict the user's click-through rate on the candidate products, so as to make information recommendations based on the click-through rate of the candidate products.

8. A computer-readable storage medium, characterized in that, The storage medium stores a computer program, which, when executed by a processor, implements the method described in any one of claims 1 to 6.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the method described in any one of claims 1 to 6.

Images