Information Recommendation Method, Apparatus, Electronic Device and Readable Storage Medium

Abstract

Embodiments of the present disclosure provide an information recommendation method, apparatus, electronic device, and readable storage medium. The method includes: obtaining a candidate information combination corresponding to a recommendation request; inputting the user information corresponding to the recommendation request and the candidate information combination into at least one sub-model to output recommendation probabilities of the candidate information combination corresponding to different behavior categories through the at least one sub-model; the at least one sub-model is trained according to historical behavior data of different categories of users; inputting the respective recommendation probabilities output by the at least one sub-model into a fusion model to output a fusion probability corresponding to the candidate information combination; and determining a target combination from the candidate information combinations according to the fusion probability. Embodiments of the present disclosure can recommend suitable dish combinations to users, reduce the problem of long decision-making time during the user's meal ordering process, not only facilitate the user's meal ordering, but also improve the accuracy of information recommendation.

Description

Technical Field

[0001] Embodiments of the present disclosure relate to the field of network technologies, and in particular, to an information recommendation method, apparatus, electronic device, and readable storage medium. Background Art

[0002] With the rapid development of informatization, the information provided by the Internet to users has grown explosively, and the needs of users have also increased day by day. How to enable users to obtain the required information in a timely and accurate manner from a vast amount of information has become an urgent problem to be solved.

[0003] Currently, an information recommendation system can collect and analyze a user's information needs, interests, usage habits, access history, etc., and screen and sort the information on the Internet according to the analysis results, so as to recommend information that meets the user's personalized needs to the user.

[0004] Existing information recommendation systems usually recommend product information, and can estimate the order probability of a user for a single product to determine the product information recommended to the user. However, for the emerging takeaway ordering industry, since a user's order may include multiple different dishes, and a huge amount of combined information can be formed between different merchants and numerous dishes, it is difficult for existing recommendation systems to estimate the probability of a user placing an order for multiple dishes at the same time, resulting in a low accuracy of dish recommendation. Summary of the Invention

[0005] Embodiments of the present disclosure provide an information recommendation method, apparatus, electronic device, and readable storage medium to improve the accuracy of information recommendation.

[0006] According to a first aspect of an embodiment of the present disclosure, an information recommendation method is provided, the method including:

[0007] Obtain a candidate information combination corresponding to a recommendation request;

[0008] Input the user information corresponding to the recommendation request and the candidate information combination into at least one sub-model, so as to output the recommendation probabilities of the candidate information combination corresponding to different behavior categories through the at least one sub-model; the at least one sub-model is trained according to different categories of historical behavior data of users;

[0009] Input the respective recommendation probabilities output by the at least one sub-model into a fusion model to output a fusion probability corresponding to the candidate information combination;

[0010] Determine a target combination from the candidate information combinations according to the fusion probability.

[0011] According to a second aspect of an embodiment of the present disclosure, an information recommendation apparatus is provided, the apparatus including:

[0012] A candidate acquisition module, configured to acquire a candidate information combination corresponding to a recommendation request;

[0013] A first prediction module, configured to input the user information corresponding to the recommendation request and the candidate information combination into at least one sub-model, so as to output recommendation probabilities corresponding to different behavior categories of the candidate information combination through the at least one sub-model; the at least one sub-model is trained according to historical behavior data of different categories of users;

[0014] A second prediction module, configured to input the respective recommendation probabilities output by the at least one sub-model into a fusion model, so as to output a fusion probability corresponding to the candidate information combination;

[0015] A target determination module, configured to determine a target combination from the candidate information combinations according to the fusion probability.

[0016] According to a third aspect of an embodiment of the present disclosure, there is provided an electronic device, including:

[0017] A processor, a memory, and a computer program stored on the memory and executable on the processor, wherein when the processor executes the program, the foregoing information recommendation method is implemented.

[0018] According to a fourth aspect of an embodiment of the present disclosure, there is provided a readable storage medium, when instructions in the storage medium are executed by a processor of an electronic device, enabling the electronic device to execute the foregoing information recommendation method.

[0019] Embodiments of the present disclosure provide an information recommendation method, apparatus, electronic device, and readable storage medium. The method includes:

[0020] By obtaining a combination of candidate information corresponding to a recommendation request, and inputting the user information corresponding to the recommendation request and the combination of candidate information into at least one sub-model to output the recommendation probabilities of different behavior categories corresponding to the combination of candidate information through the at least one sub-model; and inputting the respective recommendation probabilities output by the at least one sub-model into a fusion model to output the fusion probability corresponding to the combination of candidate information, and then the target combination can be determined from the combination of candidate information according to the fusion probability. Since the at least one sub-model is trained based on historical behavior data of different categories of users, therefore, each sub-model can reflect the recommendation probability of the combination of candidate information from a certain specific perspective, and then the fusion model fuses the recommendation probabilities output by each sub-model to obtain the fusion probability, which can comprehensively reflect the recommendation probabilities from multiple perspectives, such as the merchant perspective, the dish perspective, etc. Thus, the recommendation system can be extended from dish recommendation to dish combination recommendation. By recommending a more suitable dish combination to the user, the problem of long decision-making time during the user's meal ordering process can be reduced, which not only provides convenience for the user's meal ordering, but also can improve the accuracy of information recommendation. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the technical solutions of the embodiments of the present disclosure, the accompanying drawings required for the description of the embodiments of the present disclosure will be briefly introduced below. Obviously, the accompanying drawings in the following description are only some embodiments of the embodiments of the present disclosure. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0022] Figure 1 The flowchart of the steps of the information recommendation method in an embodiment of the present disclosure is shown;

[0023] Figure 2 The structure diagram of the information recommendation device in an embodiment of the present disclosure is shown;

[0024] Figure 3 The structure diagram of the electronic device provided in an embodiment of the present disclosure is shown. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0025] The technical solutions in the embodiments of the present disclosure will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present disclosure. Obviously, the described embodiments are some, but not all, of the embodiments of the present disclosure. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present disclosure without creative efforts belong to the scope of protection of the embodiments of the present disclosure.

[0026] Embodiment 1

[0027] Refer toFigure 1 , which shows a flowchart of the steps of an information recommendation method in an embodiment of the present disclosure, including:

[0028] Step 101, obtaining a candidate information combination corresponding to a recommendation request;

[0029] Step 102, inputting the user information corresponding to the recommendation request and the candidate information combination into at least one sub-model to output the recommendation probabilities of the candidate information combination corresponding to different behavior categories through the at least one sub-model; the at least one sub-model is trained according to historical behavior data of different categories of users;

[0030] Step 103, inputting the respective recommendation probabilities output by the at least one sub-model into a fusion model to output a fusion probability corresponding to the candidate information combination;

[0031] Step 104, determining a target combination from the candidate information combinations according to the fusion probability.

[0032] The information recommendation method of the present disclosure can be applied to a terminal, and the terminal specifically includes but is not limited to: smart phones, tablet computers, e-book readers, MP3 (Moving Picture Experts Group Audio Layer III) players, MP4 (Moving Picture Experts Group Audio Layer IV) players, laptop computers, in-vehicle computers, desktop computers, set-top boxes, smart TVs, wearable devices, and so on.

[0033] It should be noted that the information in the present disclosure includes any information that a user can obtain through a terminal, which can be commodity information, merchant information, or dish information, or can also be news, entertainment, or any other information. The embodiments of the present disclosure mainly illustrate the dish information in the takeout ordering scenario, and the information processing processes in other application scenarios can be referred to each other.

[0034] Generally, there can be multiple candidate dish combinations, and the embodiments of the present disclosure can construct a candidate dish combination set, which contains each candidate dish combination. Taking the takeout ordering scenario as an example, the candidate information combination refers to a set composed of all dish combinations that meet the delivery conditions of the location where the user's recommendation request is located and meet the user's one-time order requirements. For example: {Yuxiang shredded pork + Mapo tofu + rice, set meal 1, set meal 2, set meal 3}. Among them, Yuxiang shredded pork + Mapo tofu + rice, set meal 1, and set meal 2 are dish combinations provided by merchant A, and set meal 3 is a dish combination provided by merchant B.

[0035] Among them, the candidate information combination can be the dish combination in the packages provided by merchants, such as Package 1 of Merchant A, Package 2 of Merchant A, and Package 3 of Merchant B, or it can be the dish combination obtained by matching the dishes provided by merchants according to the user's historical order behavior, such as "Yu-Shiang Shredded Pork + Mapo Tofu + Rice" of Merchant A.

[0036] For the takeaway ordering scenario, a user's order may include multiple different dishes, and a huge amount of combined information can be formed between different merchants and numerous dishes. To improve the accuracy of recommended dish combinations, the present disclosure trains sub-models corresponding to different behavior categories based on the user's historical behavior data of different categories, and a fusion model is trained based on each sub-model. During the information recommendation process, each sub-model and the fusion model can be loaded. Each sub-model predicts the candidate information combination according to the user's recommendation request and outputs the recommendation probability that conforms to different behavior characteristics (that is, the order placement probability of the candidate information combination under different behavior characteristics), and then inputs each recommendation probability into the fusion model for secondary prediction to obtain the fusion probability. The fusion probability can represent the comprehensive order placement probability of the user for the candidate information combination. Furthermore, the target combination can be determined from the candidate information combination according to the fusion probability, and the target combination is recommended to the user.

[0037] Since the fusion probability synthesizes the recommendation probabilities of different behavior characteristics output by at least one sub-model, and one recommendation probability can reflect the order placement probability of the user for the candidate information combination from a certain specific perspective, the fusion probability can comprehensively reflect the order placement probabilities from multiple perspectives, such as the merchant perspective, the dish perspective, etc., so as to realize the expansion from dish recommendation to dish combination recommendation. By recommending a more suitable dish combination to the user, the problem of the user having a long decision-making time during the ordering process can be reduced, which provides convenience for the user to order food.

[0038] In an alternative embodiment of the present disclosure, the at least one sub-model may specifically include at least one of a merchant sub-model, a dish sub-model, and a label sub-model;

[0039] The merchant sub-model is used to output the order placement probability of the user for the merchant; the merchant sub-model is trained based on the historical behavior data generated by the user's trigger operation for merchant information;

[0040] The dish sub-model is used to output the order placement probability of the user for the dish; the dish sub-model is trained based on the historical behavior data generated by the user's trigger operation for dish information;

[0041] The label sub-model is used to output the order placement probability of the user for the label; the label sub-model is trained based on the historical behavior data generated by the user's triggering operations for merchant information, the historical behavior data generated by the user's triggering operations for dish information, the merchant labels extracted from the merchant information, and the dish labels extracted from the dish information.

[0042] In an alternative embodiment of the present disclosure, the triggering operation may at least include any one of the following: click, favorite, order. Specifically, the triggering operation may include the user's click, favorite, and order behaviors for the merchant; the user's click, favorite, and order behaviors for the dish category (such as Sichuan cuisine, hot pot, buffet, etc.); the user's click, favorite, and order behaviors for the dish, etc.

[0043] Among them, the merchant sub-model can reflect the order placement probability of the user for the candidate information combination from the merchant's perspective. For example, if a user often places orders at merchant A and rarely at merchant B, then inputting the user information and the above candidate information combination into the merchant sub-model, the recommended probability of the dish combination of merchant A may be higher than that of the dish combination of merchant B. For example, the recommended probabilities of shredded pork with fish flavor + mapo tofu + rice, set meal 1, and set meal 2 of merchant A are all higher than the recommended probability of set meal 3 of merchant B.

[0044] The dish sub-model can reflect the order placement probability of the user for the candidate information combination from the dish's perspective. For example, if a user often places an order for set meal 1 of merchant A, then inputting the user information and the above candidate information combination into the dish sub-model, the recommended probability of set meal 1 of merchant A may be higher than that of other dish combinations.

[0045] Similarly, the label sub-model can reflect the order placement probability of the user for the candidate information combination from the label's perspective. The label may include: dish labels, such as taste, ingredients, etc.; merchant labels, such as cuisine type, merchant address, etc.

[0046] In an application example of the present disclosure, taking the takeout ordering scenario as an example, it is assumed that when it is detected that the user opens the combined recommendation page in Wangjing, it is determined that a recommendation request from the user is received. A candidate information combination set, that is, a candidate dish combination set, can be constructed according to the recommendation request. Suppose it is {cold noodle set meal, hamburger set meal, small bowl dish set meal}. Then, the user information of the user and each candidate dish combination in the candidate dish combination set are respectively input into the merchant sub-model and the dish sub-model. Suppose the recommended probabilities predicted by the merchant sub-model and the dish sub-model for the cold noodle set meal are 0.3 and 0.5 respectively, the recommended probabilities predicted by the merchant sub-model and the dish sub-model for the hamburger set meal are 0.7 and 0.8 respectively, and the recommended probabilities predicted by the merchant sub-model and the dish sub-model for the small bowl dish set meal are 0.5 and 0.3 respectively. Next, the recommended probabilities of the above three set meals in the two sub-models are used as features and input into the fusion model respectively to output the fusion probability. For example, 0.3 and 0.5 of the cold noodle set meal are used as features and input into the fusion model, and a fusion probability of 0.6 is output; 0.7 and 0.8 of the hamburger set meal are used as features and input into the fusion model, and a fusion probability of 0.9 is output; and 0.5 and 0.3 of the small bowl dish set meal are used as features and input into the fusion model, and a fusion probability of 0.7 is output. Finally, sorting in descending order according to the fusion probability, the sorting result is obtained as: hamburger set meal, small bowl dish set meal, cold noodle set meal. The target combination recommended to the user can be determined according to the fusion probability. For example, the hamburger set meal is recommended to the user as the target combination.

[0047] Thus, the target combination finally recommended to the user is the result of integrating the historical behavior characteristics of the user for merchants and the historical behavior characteristics of the user for dishes, making the recommended target combination more in line with the multi-faceted needs of the user and improving the accuracy of the recommendation.

[0048] In an alternative embodiment of the present disclosure, the candidate information combination may be a candidate dish combination. The obtaining of the candidate information combination corresponding to the recommendation request may specifically include:

[0049] Step S11: Obtain the user information corresponding to the recommendation request, where the user information includes: user identifier, user's current location;

[0050] Step S12: Determine the historical behavior data generated by the trigger operation of the user identifier for dish information;

[0051] Step S13: Determine candidate merchants whose distance from the user's current location is less than a preset distance, and the dish information provided by the candidate merchants;

[0052] Step S14: Determine the candidate dish combination corresponding to the recommendation request based on the dish information provided by the candidate merchant, the historical behavior data generated by the triggering operation of the user identifier for the dish information, and the dish combination order information of the candidate merchant.

[0053] A candidate dish combination refers to a dish combination that is obtained according to user information such as the user's current location and user identifier when receiving a user's recommendation request, and that meets the delivery conditions of the user's current location, conforms to the user's historical behavior habits, and can be ordered by the user at one time.

[0054] Specifically, first obtain the user information corresponding to the recommendation request. The user information may include: user identifier, user's current location. Based on the user's current location, candidate merchants whose distance from the user's current location is less than a preset distance can be determined, as well as the dish information provided by the candidate merchants. The dish information provided by the candidate merchants can meet the delivery conditions of the user's current location. Therefore, a dish combination that conforms to the user's historical behavior habits and can be ordered by the user at one time can be determined from the dish information provided by the candidate merchants, and then the candidate dish combination can be obtained.

[0055] In the embodiments of the present disclosure, the dish combination is recommended to the user as a whole. The difference from recommending a single commodity is that the number of different combinations of different dishes provided by different candidate merchants is extremely large after any combination, and many combinations cannot meet the user's need to place an order at one time. For example, if the combination only includes "steamed bun + rice", after the user places an order for this combination, other dishes still need to be ordered separately. Therefore, the dish combination "steamed bun + rice" cannot meet the user's need to place an order at one time, while the dish combination "rice + scrambled eggs with tomatoes" can meet the user's need to place an order at one time.

[0056] In the embodiments of the present disclosure, the candidate dish combination corresponding to the recommendation request can be obtained through multiple sources, so that the candidate dish combination can meet the user's need to place an order at one time. Specifically, the candidate dish combination corresponding to the recommendation request can be determined based on the dish information provided by the candidate merchant, the historical behavior data generated by the triggering operation of the user identifier for the dish information, and the dish combination order information of the candidate merchant.

[0057] Among them, multiple sources refer to the sources for obtaining the candidate dish combination. Specifically, it may include the completed order information in the user's historical behavior data, the dish combination order information of the candidate merchant, and the dish information provided by the candidate merchant. Of course, it may also include other users' completed order information, package information matched by intelligent algorithms, etc. It can be understood that the embodiments of the present disclosure do not limit the sources for obtaining the candidate dish combination.

[0058] In an alternative embodiment of the present disclosure, the sub-model can be trained through the following steps:

[0059] Step S21: Classify the historical behavior data of the collected users to obtain historical behavior data sets of different categories; wherein, each historical behavior data set corresponds to one category;

[0060] Step S22: Sort the historical behavior data in the historical behavior data set in chronological order;

[0061] Step S23: Divide the sorted historical behavior data set in the manner of a sliding window, use the historical behavior data within the sliding window as the training set, and use the historical behavior data outside the sliding window as the prediction set;

[0062] Step S24: Train sub-models of different categories according to the training set and prediction set in the historical behavior data set corresponding to each category.

[0063] First, obtain the historical behavior data (HistoryData) of the user. The historical behavior data may include: historical behavior data generated by trigger operations such as the user's clicks, collections, comments, and orders on merchants / dishes / labels, etc. Embodiments of the present disclosure can also mark the historical behavior data to mark the order status of the user. For example, 0 indicates that the user has not placed an order, and 1 indicates that the user has placed an order.

[0064] Classify the marked historical behavior data to obtain historical behavior data sets of different categories, such as the historical behavior data set of the user for merchants, the historical behavior data set of the user for dishes, the historical behavior data set of the user for labels, etc. Among them, each historical behavior data set corresponds to one category, that is, the number of categories corresponds to the number of sub-models.

[0065] Then, sort the historical behavior data in the historical behavior data set in chronological order, divide the sorted historical behavior data set in the manner of a sliding window, use the historical behavior data within the sliding window as the training set, and use the historical behavior data outside the sliding window as the prediction set. It can be understood that the window size of the sliding window in the embodiments of the present disclosure is not limited. Specifically, extract features from the data in the training set to obtain the training data (TrainData) for training the sub-model. The prediction set (TestData) is used to verify the trained sub-model to adjust and optimize the parameters of the sub-model.

[0066] Referring to Table 1, a specific illustration of the historical behavior data of a user in the present disclosure is shown.

[0067] Table 1

[0068]

[0069] As shown in Table 1, the historical behavior data is divided into two historical behavior data sets of "user - merchant" and "user - dish" categories. Among them, "user - merchant" refers to the historical behavior data generated by the user's trigger operations on merchants, and "user - dish" refers to the historical behavior data generated by the user's trigger operations on dishes. Specifically, the two sorted historical behavior data sets are {Data 1, Data 3, Data 4, Data 2} and {Data 2, Data 4, Data 5, Data 3} respectively, which can be used to train sub - models of two different behavior categories. For these two historical behavior data sets, they are divided in the way of a sliding window. The historical behavior data within the sliding window is used as the training set, and the historical behavior data outside the sliding window is used as the prediction set. Referring to Table 2, a specific illustration of a divided training set and prediction set of the present disclosure is shown.

[0070] Table 2

[0071]

[0072] As shown in Table 2, the window size of the sliding window is 3 days. The historical behavior data within the sliding window (from April 1st to April 3rd) is used as the training set, and the historical behavior data outside the sliding window (April 4th) is used as the prediction set. In specific applications, the time point of the prediction set should be after the time point of the training set. In Table 2, the time point of the prediction set is 1 day after the time point of the training set. Of course, it can also be 1 hour later, 30 minutes later, etc.

[0073] Next, according to the training sets and prediction sets in the historical behavior data sets corresponding to each category, sub - models of different categories are trained. Specifically, according to the data features and order placement identifiers of the training set corresponding to the user's historical behavior towards merchants, a merchant sub - model is trained; according to the data features and order placement identifiers of the training set corresponding to the user's historical behavior towards dishes, a dish sub - model is trained.

[0074] It can be understood that the above - mentioned training of sub - models using two historical behavior categories is only an application example of the present disclosure, and the embodiments of the present disclosure do not limit the number and types of historical behavior categories used.

[0075] For example, according to the historical behavior data sets corresponding to n categories, the following n sub - models can be trained: subModel1, subModel2, ……, subModel i 、……、subModel n ,where subModel i =Train i (TrainData). Where Traini Represents the training process of the i-th sub-model.

[0076] Finally, use the trained sub-model to predict the prediction set, and obtain the predicted recommendation probabilities of the user for the dish combinations corresponding to the prediction set as follows: score1, score2,..., score i ,..., score n , where score i = subModel i (TestData). Among them, subModel i Represents the prediction process of the i-th sub-model. According to the predicted recommendation probabilities output by the prediction set, verify whether the sub-model reaches the optimal, adjust the parameters of the sub-model, optimize the sub-model, and obtain the optimized sub-model.

[0077] In an alternative embodiment of the present disclosure, the method may further include: using the prediction set in the historical behavior data set corresponding to each category as the training data of the fusion model, and training to obtain the fusion model.

[0078] In the embodiments of the present disclosure, each sub-model can reflect the recommendation probability of the candidate information combination from a single perspective, while the fusion model can fuse the output results of each sub-model and reflect the recommendation probability of the candidate information combination from a comprehensive perspective. Therefore, in the embodiments of the present disclosure, the prediction sets of the sub-models can be spliced together to obtain the training data of the fusion model, because the input features of the fusion model are the outputs of each sub-model. Specifically, score i can be used as the feature vector to train the fusion model (Model), that is, Model = Train(score1, score2,..., score i ,..., score n ).

[0079] In summary, by obtaining the candidate information combinations corresponding to the recommendation requests, and inputting the user information corresponding to the recommendation requests and the candidate information combinations into at least one sub-model, so as to output the recommendation probabilities of the candidate information combinations corresponding to different behavior categories through the at least one sub-model; and inputting the respective recommendation probabilities output by the at least one sub-model into a fusion model to output the fusion probability corresponding to the candidate information combination, and then the target combination can be determined from the candidate information combinations according to the fusion probability. Since the at least one sub-model is trained based on the historical behavior data of different categories of users, each sub-model can reflect the recommendation probability of the candidate information combination from a certain specific perspective, and then the fusion model fuses the recommendation probabilities output by each sub-model to obtain the fusion probability, which can comprehensively reflect the recommendation probabilities from multiple perspectives, such as the merchant perspective, the dish perspective, etc. Thus, the recommendation system can be extended from dish recommendation to dish combination recommendation. By recommending more suitable dish combinations to users, the problem of long decision-making time during the user's meal ordering process can be reduced, which not only provides convenience for the user to order meals, but also improves the accuracy of information recommendation.

[0080] Embodiment 2

[0081] Referring to Figure 2 , which shows the structural diagram of the information recommendation device in an embodiment of the present disclosure, specifically as follows.

[0082] The candidate acquisition module 201 is configured to acquire candidate information combinations corresponding to the recommendation requests;

[0083] The first prediction module 202 is configured to input the user information corresponding to the recommendation requests and the candidate information combinations into at least one sub-model, so as to output the recommendation probabilities of the candidate information combinations corresponding to different behavior categories through the at least one sub-model; the at least one sub-model is trained based on the historical behavior data of different categories of users;

[0084] The second prediction module 203 is configured to input the respective recommendation probabilities output by the at least one sub-model into a fusion model to output the fusion probability corresponding to the candidate information combination;

[0085] The target determination module 204 is configured to determine the target combination from the candidate information combinations according to the fusion probability.

[0086] Optionally, the candidate information combination is a candidate dish combination, and the information acquisition module includes:

[0087] The information acquisition sub-module is configured to acquire the user information corresponding to the recommendation requests, and the user information includes: user identification, user's current location;

[0088] The first determination sub-module is configured to determine the historical behavior data generated by the trigger operation of the user identifier for the dish information;

[0089] The second determination sub-module is configured to determine candidate merchants whose distance from the current location of the user is less than a preset distance, and the dish information provided by the candidate merchants;

[0090] The third determination sub-module is configured to determine the candidate dish combination corresponding to the recommendation request according to the dish information provided by the candidate merchants, the historical behavior data generated by the trigger operation of the user identifier for the dish information, and the dish combination order information of the candidate merchants.

[0091] Optionally, the at least one sub-model includes at least one of: a merchant sub-model, a dish sub-model, and a label sub-model;

[0092] The merchant sub-model is configured to output the order placement probability of the user for the merchant; the merchant sub-model is trained according to the historical behavior data generated by the trigger operation of the user for the merchant information;

[0093] The dish sub-model is configured to output the order placement probability of the user for the dish; the dish sub-model is trained according to the historical behavior data generated by the trigger operation of the user for the dish information;

[0094] The label sub-model is configured to output the order placement probability of the user for the dish; the label sub-model is trained according to the historical behavior data generated by the trigger operation of the user for the merchant information, the historical behavior data generated by the trigger operation of the user for the dish information, the merchant labels extracted from the merchant information, and the dish labels extracted from the dish information.

[0095] Optionally, the trigger operation includes at least any one of the following: click, favorite, and place an order.

[0096] Optionally, the device further includes: a first training module configured to train at least one sub-model; the first training module includes:

[0097] The classification sub-module is configured to classify the collected historical behavior data of the user to obtain historical behavior data sets of different categories; wherein, each historical behavior data set corresponds to one category;

[0098] The sorting sub-module is configured to sort the historical behavior data in the historical behavior data set in chronological order;

[0099] The partitioning sub-module is configured to partition the sorted historical behavior data set in a sliding window manner, take the historical behavior data within the sliding window as the training set, and take the historical behavior data outside the sliding window as the prediction set;

[0100] A training sub-module, configured to train sub-models of different categories according to the training set and the prediction set in the historical behavior data set corresponding to each category.

[0101] Optionally, the device further includes:

[0102] A second training module, configured to use the prediction set in the historical behavior data set corresponding to each category as the training data of the fusion model, and train to obtain the fusion model.

[0103] In summary, the embodiments of the present disclosure provide an information recommendation device, the device includes: a candidate acquisition module 201, configured to acquire a candidate information combination corresponding to a recommendation request; a first prediction module 202, configured to input the user information corresponding to the recommendation request and the candidate information combination into at least one sub-model, so as to output the recommendation probabilities of the candidate information combination corresponding to different behavior categories through the at least one sub-model; the at least one sub-model is trained according to the historical behavior data of different categories of users; a second prediction module 203, configured to input the respective recommendation probabilities output by the at least one sub-model into the fusion model, so as to output the fusion probability corresponding to the candidate information combination; a target determination module 204, configured to determine a target combination in the candidate information combination according to the fusion probability. The embodiments of the present disclosure can reduce the problem of long decision-making time during the user's meal ordering process, not only provide convenience for the user to order meals, but also improve the accuracy of information recommendation.

[0104] The embodiments of the present disclosure further provide an electronic device, see Figure 3 , including: a processor 301, a memory 302, and a computer program 3021 stored on the memory and executable on the processor, and when the processor executes the program, it implements the information recommendation method of the foregoing embodiments.

[0105] The embodiments of the present disclosure further provide a readable storage medium, when the instructions in the storage medium are executed by the processor of the electronic device, the electronic device can execute the information recommendation method of the foregoing embodiments.

[0106] For the device embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and for the related parts, refer to the partial description of the method embodiments.

[0107] The algorithms and displays provided herein are not inherently related to any particular computer, virtual system, or other device. A variety of general-purpose systems may also be used in conjunction with the teachings presented herein. The structure required to construct such systems will be apparent from the above description. Additionally, embodiments of the present disclosure are not directed to any particular programming language. It should be understood that the embodiments of the present disclosure described herein can be implemented using a variety of programming languages, and the descriptions of specific languages above are for the purpose of disclosing the best mode of the embodiments of the present disclosure.

[0108] In the specification provided herein, numerous specific details are set forth. However, it can be understood that embodiments of the present disclosure may be practiced without these specific details. In some instances, well-known methods, structures, and techniques have not been shown in detail so as not to obscure the understanding of this specification.

[0109] Similarly, it should be understood that, in order to streamline the present disclosure and assist in understanding one or more of the various inventive aspects, in the above description of the exemplary embodiments of the present disclosure, the various features of the embodiments of the present disclosure are sometimes grouped together into a single embodiment, figure, or description thereof. However, the methods of the present disclosure should not be construed as reflecting an intention that the claimed embodiments of the present disclosure require more features than are expressly recited in each claim. Rather, as reflected in the following claims, the inventive aspects lie in less than all of the features of the single foregoing disclosed embodiment. Thus, the claims following the detailed description are hereby expressly incorporated into the detailed description, with each claim standing on its own as a separate embodiment of the present disclosure.

[0110] Those skilled in the art will appreciate that the modules in the devices in the embodiments can be adaptively changed and disposed in one or more devices different from the embodiments. The modules or units or components in the embodiments can be combined into one module or unit or component, and in addition, they can be divided into multiple sub-modules or sub-units or sub-components. Except that at least some of such features and / or processes or units are mutually exclusive, any combination can be used to combine all the features disclosed in this specification (including the accompanying claims, abstract, and drawings) and all the processes or units of any method or device so disclosed. Unless otherwise expressly stated, each feature disclosed in this specification (including the accompanying claims, abstract, and drawings) can be replaced by an alternative feature that provides the same, equivalent, or similar purpose.

[0111] Each component embodiment of the embodiments of the present disclosure may be implemented in hardware, or in software modules running on one or more processors, or in a combination thereof. Those skilled in the art should understand that a microprocessor or a digital signal processor (DSP) may be used in practice to implement some or all of the functions of some or all of the components in the sorting device according to the embodiments of the present disclosure. The embodiments of the present disclosure may also be implemented as a device or apparatus program for performing some or all of the methods described herein. Such a program implementing the embodiments of the present disclosure may be stored on a computer-readable medium, or may be in the form of one or more signals. Such signals may be downloaded from an Internet website, or may be provided on a carrier signal, or in any other form.

[0112] It should be noted that the above embodiments illustrate the embodiments of the present disclosure rather than limit the embodiments of the present disclosure, and those skilled in the art can design alternative embodiments without departing from the scope of the appended claims. In the claims, any reference signs placed between parentheses shall not be construed as limiting the claim. The word "comprising" does not exclude the presence of elements or steps not listed in the claim. The word "a" or "an" preceding an element does not exclude the presence of a plurality of such elements. The embodiments of the present disclosure can be implemented by means of hardware including several different elements and by means of a suitably programmed computer. In the unit claims listing several devices, several of these devices may be embodied by the same item of hardware. The use of the words first, second, and third, etc. does not denote any order. These words may be interpreted as names.

[0113] Those skilled in the art can clearly understand that for the convenience and brevity of description, the specific working processes of the systems, devices, and units described above may refer to the corresponding processes in the foregoing method embodiments and will not be repeated here.

[0114] The above are only the preferred embodiments of the embodiments of the present disclosure and are not intended to limit the embodiments of the present disclosure. Any modifications, equivalent replacements, and improvements made within the spirit and principle of the embodiments of the present disclosure shall be included in the protection scope of the embodiments of the present disclosure.

[0115] The above is only the specific implementation manner of the embodiments of the present disclosure, but the protection scope of the embodiments of the present disclosure is not limited thereto. Any person skilled in the art can easily think of changes or replacements within the technical scope disclosed by the embodiments of the present disclosure, and all of them should be covered by the protection scope of the embodiments of the present disclosure. Therefore, the protection scope of the embodiments of the present disclosure shall be subject to the protection scope of the claims.

Claims

1. An information recommendation method, characterized in that, The method includes: Obtaining a candidate information combination corresponding to a recommendation request; the candidate information combination includes dish combinations of multiple merchants; Inputting the user information corresponding to the recommendation request and the candidate information combination into at least one sub-model to output, through the at least one sub-model, the recommendation probabilities corresponding to different behavior categories of the candidate information combination, where the behavior categories are obtained by dividing based on historical behavior data; the at least one sub-model is trained based on different categories of historical behavior data of users; the at least one sub-model includes at least one of a merchant sub-model, a dish sub-model, and a label sub-model; the merchant sub-model is used to output the order placement probability of a user for a merchant; the merchant sub-model is trained based on historical behavior data generated by a user's trigger operation for merchant information; the dish sub-model is used to output the order placement probability of a user for a dish; the dish sub-model is trained based on historical behavior data generated by a user's trigger operation for dish information; the label sub-model is used to output the order placement probability of a user for a dish; the label sub-model is trained based on historical behavior data generated by a user's trigger operation for merchant information, historical behavior data generated by a user's trigger operation for dish information, merchant labels extracted from the merchant information, and dish labels extracted from the dish information; the dish labels include at least one of taste and ingredients, and the merchant labels include at least one of cuisine type and merchant address; Inputting the respective recommendation probabilities output by the at least one sub-model into a fusion model to output the fusion probability corresponding to the candidate information combination; Determining a target combination from the candidate information combination according to the fusion probability.

2. The method according to claim 1, wherein The candidate information combination is a candidate dish combination, and obtaining the candidate information combination corresponding to the recommendation request includes: Obtaining the user information corresponding to the recommendation request, where the user information includes: user identifier, user's current location; Determining the historical behavior data generated by the user identifier's trigger operation for dish information; Determining candidate merchants whose distance from the user's current location is less than a preset distance, and the dish information provided by the candidate merchants; Determining the candidate dish combination corresponding to the recommendation request according to the dish information provided by the candidate merchants, the historical behavior data generated by the user identifier's trigger operation for dish information, and the dish combination order information of the candidate merchants.

3. The method according to claim 1, wherein The trigger operation includes at least any one of the following: click, favorite, place an order.

4. The method according to claim 1, wherein The sub-model is trained through the following steps: Classifying the collected historical behavior data of users to obtain historical behavior data sets of different categories; where each historical behavior data set corresponds to one category; Sorting the historical behavior data in the historical behavior data set in chronological order; Dividing the sorted historical behavior data set in a sliding window manner, using the historical behavior data within the sliding window as the training set and the historical behavior data outside the sliding window as the prediction set; Train sub-models of different categories according to the training set and prediction set in the historical behavior data set corresponding to each category.

5. The method according to claim 4, characterized in that The method further includes: Use the prediction set in the historical behavior data set corresponding to each category as the training data of the fusion model, and train to obtain the fusion model.

6. An information recommendation device, characterized in that, The device includes: A candidate acquisition module, configured to acquire a candidate information combination corresponding to a recommendation request; the candidate information combination includes a dish combination of multiple merchants; A first prediction module, configured to input the user information corresponding to the recommendation request and the candidate information combination into at least one sub-model, so as to output the recommendation probabilities of the candidate information combination corresponding to different behavior categories through the at least one sub-model, where the behavior categories are obtained by dividing according to historical behavior data; the at least one sub-model is trained according to the historical behavior data of different categories of users; the at least one sub-model includes at least one of a merchant sub-model, a dish sub-model, and a label sub-model; the merchant sub-model is configured to output the order placement probability of the user for the merchant; the merchant sub-model is trained according to the historical behavior data generated by the user's trigger operation for merchant information; the dish sub-model is configured to output the order placement probability of the user for the dish; the dish sub-model is trained according to the historical behavior data generated by the user's trigger operation for dish information; the label sub-model is configured to output the order placement probability of the user for the dish; the label sub-model is trained according to the historical behavior data generated by the user's trigger operation for merchant information, the historical behavior data generated by the user's trigger operation for dish information, the merchant labels extracted from the merchant information, and the dish labels extracted from the dish information; the dish labels include at least one of taste and ingredients, and the merchant labels include at least one of cuisine and merchant address; A second prediction module, configured to input the recommendation probabilities output by the at least one sub-model into the fusion model, so as to output the fusion probability corresponding to the candidate information combination; A target determination module, configured to determine a target combination in the candidate information combination according to the fusion probability.

7. The device according to claim 6, characterized in that The candidate information combination is a candidate dish combination, and the information acquisition module includes: An information acquisition sub-module, configured to acquire the user information corresponding to the recommendation request, where the user information includes: user identifier, user's current location; A first determination sub-module, configured to determine the historical behavior data generated by the user identifier's trigger operation for dish information; A second determination sub-module, configured to determine candidate merchants whose distance from the user's current location is less than a preset distance, and the dish information provided by the candidate merchants; A third determination sub-module, configured to determine the candidate dish combination corresponding to the recommendation request according to the dish information provided by the candidate merchants, the historical behavior data generated by the user identifier's trigger operation for dish information, and the dish combination order information of the candidate merchants.

8. An electronic device, characterized in that, Includes: A processor, a memory, and a computer program stored on the memory and executable on the processor, wherein when the processor executes the program, an information recommendation method as recited in one or more of claims 1-5 is implemented.

9. A readable storage medium, characterized in that, When instructions in the storage medium are executed by a processor of an electronic device, the electronic device is enabled to execute an information recommendation method as recited in one or more of method claims 1-5.

Images