Product recommendation method, device, electronic device and readable medium
By acquiring and processing the characteristics of the target products and objects, and using machine learning models to determine potential users, we solve the accuracy and cost problems of product recommendations such as games in the existing technology, and achieve efficient user positioning and advertising delivery.
Patent Information
- Application Number
- CN201910563662.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2019-06-26
- Publication Date
- 2024-12-06
- Estimated Expiration
- 2039-08-23
AI Technical Summary
The existing technology is difficult to accurately locate potential users of products such as games, resulting in wasted new advertising resources and poor user experience for games.
By acquiring the characteristics of the target products and target objects, processing them using machine learning models, identifying potential users to achieve accurate recommendations, reducing advertising costs and improving user experience.
It has achieved accurate delivery of target products, improved usage rate and user experience, reduced recommendation costs, and is especially suitable for the promotion of new products.
Smart Images

Figure CN110363574B_ABST
Abstract
Description
Technical Field
[0001] The present disclosure relates to the field of computer and Internet technology, and in particular to a product recommendation method, device, electronic device and computer-readable storage medium. Background Art
[0002] In the early stage of a new game’s launch, game ads will be recommended to players to attract the initial players. This is the new player acquisition phase of the game. Whether or not a sufficient number of players can be acquired during the acquisition phase often determines the survival of the game. However, advertising resources are limited, so how to accurately find potential players of the game and maximize advertising revenue is the core issue of new player acquisition.
[0003] It should be noted that the information disclosed in the above background technology section is only used to enhance the understanding of the background of the present disclosure, and therefore may include information that does not constitute the prior art known to ordinary technicians in the field. Summary of the invention
[0004] In view of this, the present disclosure provides a product recommendation method and device, an electronic device and a computer-readable storage medium, which can accurately locate potential users of a target product to improve the usage rate of the target product.
[0005] Other features and advantages of the present disclosure will become apparent from the following detailed description, or may be learned in part by the practice of the present disclosure.
[0006] According to the first aspect of an embodiment of the present disclosure, a product recommendation method is proposed, which includes: obtaining target product features of a target product; obtaining object features of a target object; processing the target product features and the object features through a first machine learning model to obtain a first processing result; determining a first target object of the target product based on the first processing result to recommend the target product to the first target object.
[0007] In some exemplary embodiments, obtaining the object characteristics of the target object includes: obtaining historical product characteristics of historical products; and determining the object characteristics according to the historical product characteristics.
[0008] In some exemplary embodiments, determining the object characteristics based on the historical product characteristics includes: obtaining a historical duration of use of the historical product by the target object within a first predetermined time; determining a weight of the corresponding historical product characteristics based on the historical duration; and obtaining the object characteristics based on the historical product characteristics and their weights.
[0009] In some exemplary embodiments, the product recommendation method further includes: obtaining a first training set; and training the first machine learning model using the first training set.
[0010] In some exemplary embodiments, obtaining a first training set includes: randomly confirming a training product among the historical products and obtaining training product features of the training product; randomly determining a training object among the target object and obtaining training object features of the training object; if the training object has used the training product within a second predetermined time, generating a positive sample in the first training set based on the training history product features and the object features; if the training object has not used the training product within the second predetermined time, generating a negative sample in the first training set based on the training product features and the object features.
[0011] In some exemplary embodiments, obtaining historical product features of historical products includes: obtaining N feature vectors of the historical products, where N is a positive integer greater than or equal to 1; and determining the historical product features according to the N feature vectors.
[0012] In some exemplary embodiments, the historical product is described by at least one system, each system includes at least one subsystem, and each subsystem includes at least one option; wherein obtaining N feature vectors of the historical product includes: generating at least one-dimensional vector according to the selected options in the subsystem; and concatenating at least one-dimensional vectors of each subsystem in each system to generate a feature vector.
[0013] In some exemplary embodiments, the product recommendation method further includes: processing the object features through a second machine learning model to obtain a second processing result; determining a second target object of the target product based on the second processing result to recommend the target product to the second target object.
[0014] In some exemplary embodiments, the product recommendation method further includes: obtaining a second training set; and training a second machine learning model using the second training set.
[0015] In some exemplary embodiments, obtaining a second training set includes: obtaining a seed object of the target product and its seed object features; obtaining a non-seed object of the target product and its non-seed object features; generating positive samples in the second training set based on the seed object features, and generating negative samples in the second training set based on the non-seed object features.
[0016] According to the second aspect of the embodiment of the present disclosure, a product recommendation device is proposed, the device comprising: a product feature acquisition module, an object feature acquisition module, a first result acquisition module and a first object determination module. The product feature acquisition module is configured to acquire the target product feature of the target product; the object feature acquisition module is configured to acquire the object feature of the target object; the first result acquisition module is configured to process the target product feature and the object feature through a first machine learning model; the first object determination module is configured to determine the first target object of the target product according to the processing result of the first machine learning model, so as to recommend the target product to the first target object.
[0017] According to the third aspect of an embodiment of the present disclosure, an electronic device is proposed, which includes: one or more processors; a storage device for storing one or more programs, when the one or more programs are executed by the one or more processors, the one or more processors implement any of the product recommendation methods described above.
[0018] According to a fourth aspect of an embodiment of the present disclosure, a computer-readable storage medium is provided, on which a computer program is stored, characterized in that when the program is executed by a processor, the product recommendation method as described in any one of the above items is implemented.
[0019] According to the product recommendation method, device, electronic device and computer-readable storage medium provided by certain embodiments of the present disclosure, the target product features of the target product and the object features of the target object are processed by the first machine learning model to obtain a first processing result, so that the first target object can be determined in the target object, so as to recommend the target product to the first target object. On the one hand, it is possible to combine the features of the target product and the features of the target object to determine the first target object with a greater probability of using the target product. When recommending the target product to the first target object, the accurate delivery of the target product can be achieved, the probability of the target product being used can be increased, and the recommendation cost of the target product can be reduced. On the other hand, the seed object that has already used the target product can be used to directly determine the potential object in the target object, which is of great significance for the promotion of new products without seed users. In addition, the target product is recommended to potential users (i.e., the first target object) who are interested in it, compared with the random push to all users without purpose in the prior art, the interference of non-potential users who are not interested in the target product can be reduced, and the user experience is improved.
[0020] It is to be understood that the foregoing general description and the following detailed description are exemplary only and are not restrictive of the present disclosure. BRIEF DESCRIPTION OF THE DRAWINGS
[0021] The drawings herein are incorporated into the specification and constitute a part of the specification, illustrate embodiments consistent with the present disclosure, and together with the specification are used to explain the principles of the present disclosure. The drawings described below are only some embodiments of the present disclosure, and for ordinary technicians in this field, other drawings can also be obtained based on these drawings without creative work.
[0022] Figure 1 A schematic diagram showing an exemplary system architecture of a product recommendation method or a product recommendation device applied to an embodiment of the present disclosure.
[0023] Figure 2 It is a schematic diagram of a hierarchical user model according to the related art.
[0024] Figure 3 The present invention is a flowchart of a product recommendation method according to an embodiment of the present invention.
[0025] Figure 4 yes Figure 3 Flowchart of step S2 in an exemplary embodiment.
[0026] Figure 5 yes Figure 4 Flowchart of step S22 in an exemplary embodiment.
[0027] Figure 6 It is a flowchart of another product recommendation method according to an embodiment of the present disclosure.
[0028] Figure 7 yes Figure 6 Flowchart of step S5 in an exemplary embodiment.
[0029] Figure 8 yes Figure 4 Flowchart of step S21 in an exemplary embodiment.
[0030] Fig.9A It is a system schematic diagram used to describe a target product according to an embodiment of the present disclosure.
[0031] Fig. 9B yes Fig.9A A schematic diagram of a main category in an exemplary embodiment.
[0032] Fig.10 yes Figure 8 Step S211 in FIG. 1 is a flowchart of an exemplary embodiment.
[0033] Fig.11 It is a schematic diagram showing a method of generating product features based on N feature vectors according to an embodiment of the present disclosure.
[0034] Fig.12 The present invention is a flowchart of another product recommendation method according to an embodiment of the present invention.
[0035] Fig.13 It is a flowchart of another product recommendation method according to an embodiment of the present disclosure.
[0036] Fig.14 yes Fig.13 Flowchart of step S9 in an exemplary embodiment.
[0037] Fig.15 The diagram is a schematic diagram showing a method of recommending a target product to a potential object according to an embodiment.
[0038] Fig.16 The figure is a block diagram of a device for recommending products according to an exemplary embodiment.
[0039] Fig.17 The diagram is a structural diagram of a computer system applied to a product recommendation device according to an exemplary embodiment. DETAILED DESCRIPTION
[0040] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in many forms and should not be construed as limited to the embodiments set forth herein; rather, these embodiments are provided so that this disclosure will be comprehensive and complete and will fully convey the concepts of the example embodiments to those skilled in the art. The same reference numerals in the figures represent the same or similar parts, and thus their repeated description will be omitted.
[0041] The described features, structures or characteristics may be combined in one or more embodiments in any suitable manner. In the following description, many specific details are provided to provide a full understanding of the embodiments of the present disclosure. However, those skilled in the art will appreciate that the technical solutions of the present disclosure may be practiced while omitting one or more of the specific details, or other methods, components, devices, steps, etc. may be adopted. In other cases, known methods, devices, implementations or operations are not shown or described in detail to avoid blurring various aspects of the present disclosure.
[0042] The accompanying drawings are only schematic diagrams of the present disclosure, and the same reference numerals in the drawings represent the same or similar parts, so their repeated description will be omitted. Some block diagrams shown in the accompanying drawings do not necessarily correspond to physically or logically independent entities. These functional entities can be implemented in software form, or in one or more hardware modules or integrated circuits, or in different networks and / or processor devices and / or microcontroller devices.
[0043] The flowcharts shown in the accompanying drawings are only exemplary and do not necessarily include all the contents and steps, nor must they be executed in the order described. For example, some steps can be decomposed, and some steps can be combined or partially combined, so the actual execution order may change according to actual conditions.
[0044] In this specification, the terms "a", "an", "the", "said" and "at least one" are used to indicate the presence of one or more elements / components / etc.; the terms "comprising", "including" and "having" are used to express an open-ended inclusion and mean that additional elements / components / etc. may exist in addition to the listed elements / components / etc.; the terms "first", "second" and "third" etc. are used only as labels and are not intended to limit the quantity of their objects.
[0045] The exemplary embodiments of the present disclosure are described in detail below with reference to the accompanying drawings.
[0046] Figure 1 A schematic diagram showing an exemplary system architecture that can be applied to the product recommendation method or product recommendation device of the embodiments of the present disclosure.
[0047] like Figure 1 As shown, the system architecture 100 may include terminal devices 101, 102, 103, a network 104 and a server 105. The network 104 is used to provide a medium for communication links between the terminal devices 101, 102, 103 and the server 105. The network 104 may include various connection types, such as wired, wireless communication links or optical fiber cables, etc.
[0048] Users can use terminal devices 101, 102, 103 to interact with server 105 through network 104 to receive or send messages, etc. Terminal devices 101, 102, 103 can be various electronic devices with display screens and supporting web browsing, including but not limited to smart phones, tablet computers, laptop computers, desktop computers, etc.
[0049] The server 105 may be a server that provides various services, such as a background management server that provides support for devices operated by users using the terminal devices 101, 102, and 103. The background management server may analyze and process the received request data, and feed back the processing results to the terminal device.
[0050] The server 105 may, for example, obtain target product features of a target product; the server 105 may, for example, obtain object features of a target object; the server 105 may, for example, process the target product features and the object features through a first machine learning model to obtain a first processing result; the server 105 may, for example, determine a first target object of the target product based on the first processing result to recommend the target product to the first target object.
[0051] It should be understood that Figure 1 The number of terminal devices, networks and servers in the figure is only for illustration. The server 105 may be a physical server or may be composed of multiple servers. According to actual needs, it may have any number of terminal devices, networks and servers.
[0052] The method provided in the embodiments of the present disclosure can be processed by any electronic device with computing processing capabilities, such as the above Figure 1 The server 105 and / or terminal devices 102 and 103 in the embodiment are described by taking the server 105 as the execution subject in the following embodiment, but the present disclosure is not limited to this.
[0053] In related technologies, a hierarchical user model can be used to find potential users. Figure 2 As shown, the hierarchical user model may include multiple levels such as core users, sub-core users, and potential users.
[0054] The core user level may include users who use products of the same series or theme as the target product. For example, for a new game with the theme of the Three Kingdoms to be promoted, all users who have played the game before are core users in the core user level.
[0055] The sub-core user level can include highly active users who are willing to try new products and users who use the target product's competing products. For example, for a new game to be promoted, users who are willing to try the new game and users who have played the new game's competing products before can be sub-core users in the sub-core user level.
[0056] The potential user level may include those found in other product categories who may be interested in the new game to be promoted. For example, for a new game to be promoted, those users who may be interested in the new game identified from product categories such as literature, video, and music are potential users in the potential user level.
[0057] However, there are at least the following drawbacks in identifying potential users of new games through hierarchical user models:
[0058] 1. The number of potential users is small. The number of potential users identified through the use of products of the same series or theme is limited, and it is impossible to launch a large number of them. It is also impossible to identify enough potential users with high accuracy for product recommendations through other means.
[0059] 2. Information about users of competing products is generally difficult to obtain and has low universality.
[0060] 3. It is not conducive to the promotion of games that are not in the same or similar categories. For games that are not in the same series or similar themes, it is impossible to accurately identify potential users who are more likely to use new games.
[0061] The technical solution provided by the embodiment of the present disclosure can accurately locate a large number of potential objects. When the target product is recommended to the potential object, the utilization rate of the target product will be greatly improved.
[0062] Figure 3 It is shown according to an exemplary embodiment that the target product may refer to various physical or non-physical products such as games, literary works, videos, music, electronic products, clothes, etc. The present disclosure does not limit the types of target products, and any product that needs to be recommended may be the target product referred to in the embodiment of the present disclosure.
[0063] In the following embodiments, the target product is taken as an example of a new game to be released online, but it is not used to limit the protection scope of the present disclosure.
[0064] In some embodiments, each target product may be described by at least one system, each system may include at least one subsystem, and each subsystem may include at least one option.
[0065] In some embodiments, at least one-dimensional vector may be generated according to the selected options in the subsystems; and then the at least one-dimensional vectors of each subsystem in each system may be concatenated to generate a feature vector.
[0066] In some embodiments, N feature vectors of the target product may be obtained, where N is a positive integer greater than or equal to 1, and then the target product features may be determined based on the N feature vectors.
[0067] In step S2, object features of the target object are obtained.
[0068] In some embodiments, the target objects may refer to those objects that are initially determined to be likely to use the target product. For example, assuming that the target product is a new game, when recommending the new game, users who have played the game or registered a game account may be considered as the recommended objects first, and then these recommended objects may be the target objects of the new game. Of course, users who may be interested in the new game may also be found from users of pan-entertainment such as literature, video, music, etc. as the target objects.
[0069] In step S3, the target product features and the object features are processed by a first machine learning model to obtain a first processing result.
[0070] In some embodiments, the target product characteristics of the target product and the object characteristics of each target object can be input into the first machine learning model. After the first machine learning model processes the target product characteristics and the target object characteristics, the probability of each target object using the target product can be obtained as the first processing result; or, the judgment result of whether each target object will use the target product can be obtained as the first processing result.
[0071] In step S4, a first target object of the target product is determined according to the first processing result, so as to recommend the target product to the first target object.
[0072] In some embodiments, when the probability of each target object using the target product is determined based on the target product characteristics and the object characteristics of each target object, the first target object of the target product can be determined based on the probability corresponding to each target object. For example, a probability threshold can be set in advance (the specific size can be adjusted according to the actual application scenario, and the present disclosure is not limited thereto, for example, it can be set to 0.5), and those target objects whose probability of using the target product is greater than the probability threshold are determined as the first target object. For another example, the target object with a higher probability of using the target product can also be determined as the first target object based on the scheduled delivery volume of the target product (assuming that the scheduled delivery volume is 10,000 objects, then they can be arranged in descending order from large to small according to the probability, and the first 10,000 target objects with a higher probability are selected as the first target object. The scheduled delivery volume here is only used for example and can be set according to the specific situation).
[0073] Here, when determining the first target object based on the probability of each target object using the target product predicted by the first machine learning model, the amount of potential user data of the target product obtained is large enough to meet the predetermined delivery volume of the target product.
[0074] In other embodiments, it can also be determined whether the target object will use the target product based on the target product characteristics and the object characteristics of the target object. If the first machine learning model predicts that a target object will use the target product, then the corresponding target object can be used as the first target object.
[0075] The product recommendation method provided by the embodiment of the present disclosure processes the target product feature of the target product and the object feature of the target object through the first machine learning model to obtain the first processing result, so that the first target object can be determined in the target object, so as to recommend the target product to the first target object. On the one hand, it is possible to combine the characteristics of the target product and the characteristics of the target object to determine the first target object with a greater probability of using the target product. When recommending the target product to the first target object, the accurate delivery of the target product can be achieved, the probability of the target product being used can be increased, and the recommendation cost of the target product can be reduced. On the other hand, the seed object that has used the target product can be used, and the potential object can be directly determined in the target object, which is of great significance for the promotion of new products without seed users. In addition, the target product is recommended to potential users (i.e., the first target object) who are interested in it. Compared with the random push to all users without purpose in the prior art, the interference of non-potential users who are not interested in the target product can be reduced, and the user experience is improved.
[0076] Figure 4 yes Figure 3 Step S2 in FIG. 1 is a flow chart of an exemplary embodiment. Figure 4 As shown, in the embodiment of the present disclosure, the above step S2 may further include the following steps.
[0077] In step S21, historical product features of historical products are obtained.
[0078] In some embodiments, the historical products refer to products of the same or similar product type as the target product, and the historical products are products that have been publicly released or sold or used by users. For example, assuming that the target product is a new game that is about to be launched, the corresponding historical products include old games that have been publicly released or launched and used by users, such as historical versions of the new game, other games of the same type as the new game, or other games that are unrelated to the new game.
[0079] In step S22, the object characteristics are determined according to the historical product characteristics.
[0080] Figure 5 yes Figure 4 Step S22 in FIG. 1 is a flowchart of an exemplary embodiment. Figure 5As shown, in the embodiment of the present disclosure, the above step S22 may further include the following steps.
[0081] In step S221, the historical duration of the target object using the historical product within the first predetermined time is obtained.
[0082] In some embodiments, the first scheduled time can be specified according to the actual application scenario. For example, the period from 120 to 30 days before the current day can be specified as the first scheduled time, or the past three months can be specified as the first scheduled time. This is only used for illustration and the present disclosure does not limit this.
[0083] In some embodiments, obtaining the historical duration of the target object's use of the historical product within the first predetermined time may include: obtaining the historical products used by the target object within the first predetermined time and the historical duration of the use of the corresponding historical products. For example, the games played by each target object within the past 120 days to the past 30 days and the historical duration corresponding to each game played may be obtained.
[0084] In step S222, the weight of the corresponding historical product feature is determined according to the historical duration.
[0085] In some embodiments, the weight of the historical product feature may be determined according to the historical duration of use of the historical product by the target object within the first predetermined time.
[0086] For instance, taking games as an example, assuming that a target object has played Game A, Game B, and Game C within the first predetermined time, and the historical duration that the target object has spent on Game A, Game B, and Game C within the first predetermined time is a, b, and c hours respectively (a, b, c are all greater than or equal to 0), then for the target object, the weights corresponding to Game A, Game B, and Game C can be Ga=a / (a+b+c), Gb=b / (a+b+c), and Gc=c / (a+b+c), respectively.
[0087] In step S223, the object features are obtained according to the historical product features and their weights.
[0088] In some embodiments, the sum of the products of the historical product features of the historical products used by the target object and the corresponding weights may be used as the object feature of the target object.
[0089] For example, assuming that a target object has played Game A, Game B and Game C within the first predetermined time, and the historical product features of Game A, Game B and Game C are fa, fb and fc respectively, according to the above step S222, for this target object, the corresponding weights of Game A, Game B and Game C are Ga, Gb and Gc respectively, then for this target object, its object feature can be expressed as Ga*fa+Gb*fb+Gc*fc.
[0090] The above embodiment first obtains the historical product features of the historical product, and then generates object features according to the historical product features to map the historical product features to the target object, so as to obtain the target object's preference for the features of the historical product.
[0091] It should be noted that the calculation method of the object features of the above-mentioned target object is not limited to the above-mentioned example, and other mathematical methods can also be used to perform appropriate deformation processing on it, which is not limited in the present disclosure.
[0092] Figure 6 is a flowchart of another product recommendation method according to an embodiment of the present disclosure. Compared with the above embodiment, Figure 6 The difference between the embodiments is that the following steps may also be included.
[0093] In step S5, a first training set is obtained.
[0094] In step S6, the first machine learning model is trained using the first training set.
[0095] In some embodiments, the first machine learning model may include, for example, an XGBoost (eXtremeGradient Boosting) machine learning model, but the present disclosure is not limited thereto. The first machine learning model may also be, for example, any one or a combination of multiple of a logistic regression model, a neural network model, a deep learning model, etc.
[0096] Among them, XGBoost is one of the boosting algorithms. The idea of the boosting algorithm is to integrate many weak classifiers together to form a strong classifier. XGBoost is to continuously add trees and continuously perform feature splitting to grow a tree. Each time a tree is added, it is actually learning a new function to fit the residual of the last prediction. When the training is completed and a predetermined number of trees are obtained, the score of a sample is predicted. In fact, according to the characteristics of this sample, it will fall into a corresponding leaf node in each tree. Each leaf node corresponds to a score. Finally, you only need to add up the scores corresponding to each tree to get the predicted value of the sample. XGBoost has at least the following advantages: it uses many strategies to prevent overfitting, such as regularization terms; the objective function optimization utilizes the second-order derivative of the loss function with respect to the desired function; it supports parallelization. Although the relationship between trees is serial, nodes at the same level can be parallelized. Specifically, for a certain node, the best split point is selected within the node, and the gain of the candidate split point is calculated using multi-threaded parallelism, which makes the training speed fast; it adds processing for sparse data; cross-validation, when the prediction result is already very good, you can stop building the tree in advance to speed up the training; it supports setting sample weights, which are reflected in the first-order derivative g and the second-order derivative h. By adjusting the weights, you can pay more attention to some samples.
[0097] Figure 7 yes Figure 6 Step S5 in FIG. 1 is a flow chart in an exemplary embodiment. Figure 7 As shown, in the embodiment of the present disclosure, the above step S5 may further include the following steps.
[0098] In step S51, a training product is randomly identified from the historical products, and a training product feature of the training product is acquired.
[0099] In step S52, a training object is randomly determined among the target objects, and a training object feature of the training object is acquired.
[0100] In some embodiments, the step of acquiring the training object features is consistent with the process of acquiring the object features of the target object described above.
[0101] In step S53, if the training subject has used the training product within the second predetermined time, a positive sample in the first training set is generated according to the training product feature and the training subject feature.
[0102] In some embodiments, the second scheduled time may also be set according to actual conditions. For example, the second scheduled time may be later than the first scheduled time, and nearly 30 days may be specified as the second scheduled time, but the present disclosure is not limited thereto.
[0103] In some embodiments, if the training object has used the training product within the second predetermined time, the training product features and the training object features can be spliced to generate a positive sample in the first training set, for example, labeled "1", but the present disclosure is not limited to this.
[0104] In step S54, if the training subject has not used the training product within the second predetermined time, a negative sample in the first training set is generated according to the training product characteristics and the training subject characteristics.
[0105] In some embodiments, if the training object has not used the training product within the second predetermined time, the training product features and the training object features can be concatenated to generate negative samples in the first training set, for example, labeled "0", but the present disclosure is not limited to this.
[0106] In some embodiments, the number of negative samples in the first training set may be 3-5 times the number of positive samples, but the present disclosure is not limited thereto.
[0107] Figure 8 yes Figure 4 Step S21 in FIG. 1 is a flowchart in an exemplary embodiment. Figure 8 As shown, in the embodiment of the present disclosure, the above step S21 may further include the following steps.
[0108] In step S211, N feature vectors of the historical product are obtained, where N is a positive integer greater than or equal to 1.
[0109] In step S212, the historical product features are determined according to the N feature vectors.
[0110] In some embodiments, each historical product may be described by at least one system, each system may include at least one subsystem, and each subsystem may include at least one option.
[0111] Fig.9A is a system schematic diagram for describing a target product according to an embodiment of the present disclosure. Fig.9A In the embodiment of FIG. 1 , assuming that each historical product is a game that has been launched, the user can Fig.9AThe 23 systems shown describe each game that has been launched: PvP (Player versus Player) system (describes content related to player-to-player battles), combat system (describes content related to player battles, such as battle types and combat balance), social system (describes various relationships in the game, such as the upper limit of the number of friends of game characters, etc.), sound system (describes the sound effects of the game, such as whether the sound effects match the plot, the degree of influence of the sound effects on the game progress, etc.), plot system (describes the number of plots and side quests in the game, etc.), task system (describes whether the game has daily tasks and the time required for the tasks, etc.), trading system (describes the trading methods in the game and the types / quantities of tradable items, etc.), gang system (describes whether there are gang tasks in the game, the degree of influence of gangs on the game progress, etc.), upgrade system (describes the types of upgradeable items in the game, etc.), promotion system (describes whether there will be notifications when the game version is upgraded, etc.), achievement system (describes whether there is an achievement system in the game, etc. The following are some of the descriptions: description of the game system (description of whether there is a weather system in the game, the number of weather systems, etc.), gameplay system (description of game content, gameplay time, etc.), action system (description of whether the actions in the game are smooth), reward system (description of whether there are daily rewards in the game and the number of rewards, etc.), character system (description of whether multiple game characters can be created in the game and the number of optional characters, etc.), collection system (description of the number of items that can be collected in the game, etc.), commercialization system (description of the types and quantities of items that can be purchased in the game, etc.), control system (description of the number of control buttons in the game and the type of controller, etc.), economic system (description of the number of currency types in the game, whether the economy is open-loop, etc.), ranking system (description of the number of rankings in the game, etc.), novice system (description of the teaching time and teaching content in the game, etc.) and graphics system (description of the graphics style in the game and whether the game interface is simple, etc.), etc.
[0112] In some embodiments, each system may include at least one subsystem.
[0113] like Fig.9AAs shown, when a game that has been launched is described through a gameplay system, the gameplay system may include several subsystems, such as main category, subcategory, gameplay time consumption, attention consumption degree, difficulty to get started, and the amount of game content. Among them, the main category subsystem can describe the type of game, the subcategory subsystem can further describe the subdivision type of the game, the gameplay time consumption subsystem can describe the time consumption to complete the core content of the game, the attention consumption degree subsystem can describe the concentration level of playing the game (for example: very focused / focused / occasionally look at / mostly completed automatically), and the difficulty to get started subsystem can describe the difficulty of getting started with the game (for example: very difficult / average / you can get started just by looking at it).
[0114] In some embodiments, each subsystem may further include at least one option.
[0115] Fig. 9B yes Fig.9A Schematic diagram of the main category subsystem in an exemplary embodiment.
[0116] like Fig. 9B As shown in the figure, the main category subsystem is mainly used to describe the main types of games, among which the main category subsystem can include 19 options such as action, shooting, leisure, survival, role-playing, sports, sandbox, development, competition, fighting, puzzle, cool running, adventure, construction, game production, tower defense, simulation, strategy and others. The describer can select at least one option in each subsystem as the selected option based on his own experience or according to the scoring rules to evaluate the game that has been launched.
[0117] in, Fig. 9B The scoring rule box shows the types of some games: reaction combat games include: FPS (First Person Shooting) games COD (Call of Duty) or CF (CrossFire) games; strategy combat games include: DOTA (Defense of the Ancients); tower defense games include: Orcs Must Die; puzzle games include: Love Elimination Every Day; puzzle physics games include: pinball games include; puzzle others include: Monument Valley; rhythm games include: Rhythm Master; driving games include: Need for Speed; endless running games include: Escape from the Temple; word guessing games include: Draw and Guess; board games include: Texas Hold'em and Landlord.
[0118] It should be noted that the description dimensions of the game are not limited to the 23 systems listed above, and may include more or fewer description dimensions. At the same time, when the category of historical products changes, appropriate description methods can be used to determine the corresponding historical product features according to actual conditions.
[0119] Fig.10 yes Figure 8 Step S211 in FIG. 1 is a flowchart of an exemplary embodiment. Fig.10 As shown, in the embodiment of the present disclosure, the above step S211 may further include the following steps.
[0120] In step S2111, a vector of at least one dimension is generated according to the selected option in the subsystem.
[0121] In some embodiments, each historical product may be described in at least one system, each system may include at least one subsystem, each subsystem may include at least one option, and at least one option may be selected in each subsystem as the selected option.
[0122] In some embodiments, at least one one-dimensional vector may be generated according to the selected option in the subsystem. For example, still taking the game as an example, assuming that in the main category subsystem, if the shooting option is selected, a one-dimensional vector with a shooting option value of 1 and an action option value of 0 is generated; if the action option is selected, a one-dimensional vector with a shooting option value of 0 and an action option value of 1 is generated; if the ray option and the action option are selected at the same time, a one-dimensional vector with a shooting option value of 1 and an action option value of 1 is generated.
[0123] In step S2112, at least one-dimensional vectors of each subsystem in each system are concatenated to generate a feature vector.
[0124] In some embodiments, since each system includes at least one subsystem, a vector of at least one dimension can be generated corresponding to the selected options in each subsystem. Then, multiple vectors of at least one dimension can be generated according to the description of the historical product by each subsystem in each system, and a feature vector can be generated by sequentially concatenating the multiple vectors of at least one dimension.
[0125] Fig.11 It is a schematic diagram showing a method of generating product features based on N feature vectors according to an embodiment of the present disclosure.
[0126] like Fig.11 As shown in FIG. 1 , after the historical product description is completed through each system, the description result will be stored in the database 110. Fig.11As shown, Na represents the product name, per represents the name of the person describing it, and each subsystem will have at least one selected option corresponding to it.
[0127] In some embodiments, the person who describes the historical product may be someone who is relatively familiar with the historical product, so as to improve the accuracy of the description of the historical product.
[0128] In some embodiments, a feature vector may be generated based on the description result of each describer on the historical product. Therefore, when N (N is a positive integer greater than or equal to 1) describers describe the historical product, N feature vectors may be generated based on the N description results.
[0129] In some embodiments, vector addition and averaging may be performed on the N feature vectors to generate the historical product features of the historical product.
[0130] like Fig.11 As shown, each describer's description of the product can generate a corresponding feature vector. For example, a feature vector 112 of the target product Na can be generated based on the description 111 of the target product Na by describer per1: 0〡1…0〡0〡1…0〡1…; a feature vector 114 of the target product Na can be generated based on the description 113 of the target product Na by describer per2: 1〡0…1〡1〡0…1〡0….
[0131] In some embodiments, multiple feature vectors of the target product Na may be, for example, Fig.11 The vectors 112 and 114 shown are added and averaged to determine the historical product feature 115 of the historical product as 0.5|0.5…0.5|0.5|0.50…0.5|0.5….
[0132] It should be noted that the above only cites two eigenvectors as an example, but is not actually limited to this.
[0133] In the above embodiment, the historical product features are determined by using N feature vectors, which can avoid the problem of excessive deviation between the historical product features and the actual product features due to the error of a single feature vector.
[0134] The disclosed embodiment describes historical products through at least one system (each system includes at least one subsystem), and can describe historical products from multiple perspectives, and refine the features of historical products so that the feature vector generated by the system can describe the historical products in a more fine-grained manner.
[0135] In some embodiments, the above method for acquiring historical product features may also be applied to acquiring product features such as target product features and training product features.
[0136] Fig.12 1 is a flowchart of another product recommendation method according to an embodiment of the present disclosure. Compared with the above embodiment, the embodiment of the present disclosure is different in that it can also include the following steps.
[0137] Step S7: Process the object features through a second machine learning model to obtain a second processing result.
[0138] Step S8: determining a second target object of the target product according to the second processing result, so as to recommend the target product to the second target object.
[0139] Fig.13 1 is a flowchart of another product recommendation method according to an embodiment of the present disclosure. Compared with the above embodiment, the embodiment of the present disclosure is different in that it can also include the following steps.
[0140] In step S9, a second training set is obtained.
[0141] In step S10, the second machine learning model is trained using the second training set.
[0142] In some embodiments, the second machine learning model may also adopt an XGBoost machine learning model, but the present disclosure is not limited to this.
[0143] Fig.14 yes Fig.13 Flowchart of step S9 in an exemplary embodiment.
[0144] Step S91, obtaining the seed object of the target product and its seed object features.
[0145] In some embodiments, the seed objects of the target product include objects that have used the target product. For example, assuming that the new game to be released in the above embodiment has been released, in the early stage of the new game product being launched, some trial new objects have played the game (for example, the new game is recommended to the above first target objects, and at least some of the first target objects click and download the new game), and these trial new objects can serve as the seed objects of the new game that has been launched.
[0146] In some embodiments, the seed object features of various sub-objects may be obtained by the following steps:
[0147] Step 1: Obtain the historical products (including the target product) used by various sub-objects within the third predetermined time (assuming the last three months) and the historical duration of using the corresponding historical products. For example, assuming that the historical product is an old game that was released a long time ago, then the games played by various sub-objects in the past three months and the corresponding historical duration of playing each game in the past three months can be obtained.
[0148] Step 2: The weights of historical product features can be determined based on the historical duration of historical products used by various sub-objects within the third predetermined time. For example, assuming that a target object has played game A, game B, and game C within the first predetermined time, and the historical duration of the target object's use of game A, game B, and game C within the first predetermined time is a, b, and c hours respectively (a, b, c are all greater than or equal to 0), then for the target object, the weights corresponding to game A, game B, and game C can be G, ... a =a / (a+b+c), G b = b / (a+b+c) and G c =c / (a+b+c).
[0149] Step 3: The sum of the products of the historical product features of the historical products used by the seed object and the corresponding weights can be used as the object features.
[0150] For example, suppose a target object has played game A, game B, and game C within the first predetermined time, and the historical product features of game A, game B, and game C are f a 、f b and f c According to the above step S222, for the target object, the weights of game A, game B and game C are G and a , G b and G c , then for the target correspondence, its object feature can be expressed as G a *f a +G b *f b +G c *f c .
[0151] Step S92, obtaining the non-seed objects of the target product and their non-seed object features.
[0152] In some embodiments, the non-seed objects of the target product may refer to objects that have not used the target product.
[0153] For example, if the target product is a new game that is about to be released online, since the new game has not yet been released online, most of the users in the game disk have not played the new game. In this case, users can be randomly selected from the game disk as non-seed objects of the new game.
[0154] In some embodiments, the game disk may include users associated with the game and their related game data.
[0155] Step S93: generating positive samples in the second training set according to the seed object features, and generating negative samples in the second training set according to the non-seed object features.
[0156] For example, the seed object feature is marked as "1" and the non-seed object feature is marked as "0".
[0157] In some embodiments, the number of negative samples in the second training set may be 3-5 times the number of positive samples.
[0158] The product recommendation method provided by the embodiment of the present disclosure trains a second machine learning model based on seed object features including target product features (because the seed object has used the target product, the seed object features include the target product features) and non-seed objects, and then uses the second machine learning model to process the object features of the target object to obtain a second processing result, so that the second target object can be determined using the second processing result. That is, by spreading the seed object, the second target object with a greater probability of using the target product can be determined, and the probability of the target product being used can be increased when recommending the target product to the second target object.
[0159] Fig.15 The diagram is a schematic diagram showing a method of recommending a target product to a potential object according to an embodiment.
[0160] like Fig.15 As shown, the target product is a new game to be released online, and the target object is a user in the game market. This is illustrated by way of example, but is not intended to limit the scope of protection of the present disclosure.
[0161] In some embodiments, the game features of the new game can be first obtained, and then the object features of each user in the game disk can be obtained, and then the new game features and the object features of each user can be processed by a first machine learning model to obtain a first processing result, and finally, potential users in the game disk who are willing to try the new game can be determined based on the first processing result to recommend the new game to the potential users.
[0162] In some embodiments, each game may be described by at least one system, each system may include at least one subsystem, and each subsystem may include at least one option.
[0163] In some embodiments, the game features of a game (including new games and historical games) can be obtained by the following steps: generating at least one-dimensional vector based on the selected options in the subsystem describing the game, and then splicing at least one-dimensional vectors of each subsystem in each system to generate a feature vector of the above game; obtaining N feature vectors of the above game; and determining the game features based on the N feature vectors.
[0164] In some embodiments, the game features of each game in the game disk can be obtained according to the above method, and then the object features of each user in the game disk can be determined according to the game features of each game. For example, the object features of the user can be determined by the following steps: obtaining the games played by the user within the first predetermined time (e.g., the last 120 days to the last 30 days) and the corresponding game time of each game played within the first predetermined time; determining the weight of the game features of each game according to the game time, assuming that user c played game a for 3 hours and game b for 7 hours within the first predetermined time, then for the object features of user c, the game features of game a account for 0.3, and the game features of game b account for 0.7; the sum of the products of the game features of the games played by the user within the first predetermined time and the corresponding weights is used as the object features of the user.
[0165] In some embodiments, user U and game G may be randomly selected from the large set. If user U has played game G within the second preset time (e.g., within 30 days), the object features of user U may be combined with the game features of game G to generate a positive sample in the first training set with a label of 1; if user U has not played game G within the second preset time (e.g., within 30 days), the object features of user U may be combined with the game features of game G to generate a negative sample in the first training set with a label of 0.
[0166] In some embodiments, the first machine learning model may be trained using a first training set, and the first machine learning model may be, for example, an XGBoost machine learning model.
[0167] In some embodiments, users and their object features can be randomly extracted from the game disk and combined with new game features to generate a prediction sample of the prediction set. Then, the prediction sample generated by the new game features and the object features is processed using a first machine learning model to obtain the probability that the extracted user is willing to try the new game or the judgment result of whether the extracted user is willing to try the new game.
[0168] In some embodiments, the users in the game disk can be sorted from high to low by probability, and the first potential user who is likely to try the new game can be determined based on the scheduled delivery volume of the new game, and the game can be recommended to the first potential user in the form of an advertising link. Fig.15 As shown, the recommended user can click the early download button to enter the game interface to try to play the game.
[0169] In some embodiments, a second machine learning model may also be used to process the object features of the users in the game disk to obtain a second potential user in the game disk.
[0170] In some embodiments, the seed object features of the seed object of the new game may be obtained first, and the seed object of the new game includes users who have played the new game. Assuming that the new game in the above embodiment has been released, at the beginning of the launch of the new game product, some trial new objects have played the game (for example, the new game is recommended to the above first potential users, and at least some of the first potential users click and download the new game), these trial new objects can be used as the seed objects of the new game that has been launched.
[0171] In some embodiments, the historical games (including new games) played by the seed object within a third predetermined time (assuming it is nearly three months) and the corresponding game time of the historical games can be first obtained; then the weights of the game features of each historical game are determined according to the game time of the historical games played by various sub-objects within the third predetermined time (the process of obtaining the game features has been introduced in the above process); the sum of the products of the game features of the historical games played by the seed object and the corresponding weights is used as the seed object feature of the seed object.
[0172] In some embodiments, positive samples in the second training set may be generated according to the seed object features, and negative samples in the second training set may be generated according to the non-seed object features.
[0173] In some embodiments, the second training set can be used to train the second machine learning model, and the second machine learning model can also adopt an XGBoost machine learning model, but the present disclosure is not limited to this.
[0174] In some embodiments, the object features of the users in the game disk (the acquisition process has been described above) can be processed by a second machine learning model to obtain the probability that the user is willing to try the new game or the judgment result of whether the user is willing to try the new game.
[0175] In some embodiments, the users in the game disk can be sorted from high to low by probability, and the second potential user with a higher probability of being willing to play the new game can be determined based on the scheduled delivery volume of the new game, and the game can be recommended to the second potential user in the form of an advertising link. Fig.15 As shown, the recommended user can click the early download button to enter the game interface to try to play the game.
[0176] In the above-mentioned embodiment, not only the game features of the new game and the object features of the target object are processed by the first machine learning model to obtain the first processing result, but also the seed object features of the seed object are processed by the second machine learning model to obtain the second processing result, so that potential users who are willing to try the new game can be determined in the target object, so as to recommend the new game to the potential users. On the one hand, this method can combine the game features of the new game and the object features of the target object to determine potential users who have a greater chance of playing the new game. When recommending the new game to the potential user, the new game can be accurately delivered, the chance of being clicked on can be increased, and the recommendation cost of the new game can be reduced. On the other hand, compared with the purposeless random push to all users in the prior art, the new game is recommended to potential users who are interested in it, which can reduce the interference of non-potential users who are not interested in the new game and improve the user experience.
[0177] Fig.16 is a block diagram of a product recommendation device according to an exemplary embodiment. Fig.16 The device 160 includes a product feature acquisition module 1601, an object feature acquisition module 1602, a first result acquisition module 1603 and a first object determination module 1604.
[0178] Among them, the product feature acquisition module 1601 is configured to acquire the target product features of the target product; the object feature acquisition module 1602 is configured to acquire the object features of the target object; the first result acquisition module 1603 is configured to process the target product features and the object features through a first machine learning model; the first object determination module 1604 is configured to determine the first target object of the target product according to the processing result of the first machine learning model, so as to recommend the target product to the first target object.
[0179] In some embodiments, the object feature acquisition module 1602 may include: a historical product feature acquisition submodule and an object feature acquisition submodule.
[0180] The historical product feature acquisition submodule may be configured to acquire the historical product features of the historical products; and the object feature acquisition submodule may be configured to determine the object features according to the historical product features.
[0181] In some embodiments, the object feature acquisition submodule may include: a history duration determination unit, a weight determination unit, and an object feature determination unit.
[0182] Among them, the historical duration determination unit can be configured to obtain the historical duration of the target object using the historical product within a first predetermined duration; the weight determination unit can be configured to determine the weight of the corresponding historical product feature according to the historical duration; the object feature determination unit can be configured to obtain the object feature according to the historical product feature and its weight.
[0183] In some embodiments, the first result acquisition module 1603 may include: a first training set acquisition submodule and a first machine learning model determination submodule.
[0184] Among them, the first training set acquisition submodule can be configured to acquire a first training set; the first machine learning model determination submodule can be configured to train the first machine learning model using the first training set.
[0185] In some embodiments, the first training set acquisition submodule may include: a training product feature acquisition unit, a training object feature acquisition unit, a positive sample acquisition unit, and a negative sample acquisition unit.
[0186] Among them, the training product feature acquisition unit can be configured to randomly confirm a training product from the historical products and acquire the training product features of the training product; the training object feature acquisition unit can be configured to randomly determine a training object from the target object and acquire the training object features of the training object; the positive sample acquisition unit can be configured to generate positive samples in the first training set based on the historical training product features and the training object features if the training object has used the historical training product within the second predetermined time period; the negative sample acquisition unit can be configured to generate negative samples in the first training set based on the historical training product features and the training object features if the training object has not used the historical training product within the second predetermined time period.
[0187] In some embodiments, the historical product feature acquisition submodule may include: a feature vector acquisition unit and a historical product feature determination unit.
[0188] Among them, the feature vector acquisition unit can be configured to acquire N feature vectors of the historical product, where N is a positive integer greater than or equal to 1; the historical product feature determination unit can be configured to determine the historical product feature according to the N feature vectors.
[0189] In some embodiments, the historical product is described by at least one system, each system includes at least one subsystem, and each subsystem includes at least one option.
[0190] In some embodiments, the feature vector acquisition unit may include: a one-dimensional vector acquisition subunit and a feature vector determination subunit.
[0191] Among them, the one-dimensional vector acquisition subunit can be configured to generate at least one-dimensional vector according to the selected option in the subsystem; the feature vector determination subunit can be configured to splice at least one-dimensional vectors of each subsystem in each system to generate a feature vector.
[0192] In some embodiments, the product recommendation device 160 may further include: a second processing module and a second target object acquisition module.
[0193] Among them, the second processing module can be configured to process the object features through a second machine learning model; the second target object acquisition module can be configured to determine the second target object of the target product based on the processing results of the second machine learning model to recommend the target product to the second target object.
[0194] In some embodiments, the second target object acquisition module may include: a second training set acquisition submodule and a second machine learning model acquisition submodule.
[0195] Among them, the second training set acquisition submodule can be configured to acquire a second training set; the second machine learning model acquisition submodule can be configured to train a second machine learning model using the second training set.
[0196] In some embodiments, the second training set acquisition submodule may include: a seed object feature acquisition unit, a non-seed object acquisition unit, and a sample determination unit.
[0197] Among them, the seed object acquisition unit can be configured to acquire the seed object and its seed object features of the target product; the non-seed object acquisition unit can be configured to acquire the non-seed object and its non-seed object features of the target product; the sample determination unit can be configured to generate positive samples in the second training set according to the seed object features, and generate negative samples in the second training set according to the non-seed object features.
[0198] Since the various functional modules of the apparatus 160 of the exemplary embodiment of the present disclosure correspond to the steps of the exemplary embodiment of the above-mentioned product recommendation method, they will not be described in detail here.
[0199] Reference below Fig.17 , which shows a schematic diagram of the structure of a computer system 1700 of a terminal device suitable for implementing an embodiment of the present application. Fig.17 The terminal device shown is merely an example and should not bring any limitation to the functions and scope of use of the embodiments of the present application.
[0200] like Fig.17 As shown, the computer system 1700 includes a central processing unit (CPU) 1701, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 1702 or a program loaded from a storage part 1708 into a random access memory (RAM) 1703. In the RAM 1703, various programs and data required for the operation of the system 1700 are also stored. The CPU 1701, the ROM 1702, and the RAM 1703 are connected to each other via a bus 1704. An input / output (I / O) interface 1705 is also connected to the bus 1704.
[0201] The following components are connected to the I / O interface 1705: an input section 1706 including a keyboard, a mouse, etc.; an output section 1707 including a cathode ray tube (CRT), a liquid crystal display (LCD), etc., and a speaker, etc.; a storage section 1708 including a hard disk, etc.; and a communication section 1709 including a network interface card such as a LAN card, a modem, etc. The communication section 1709 performs communication processing via a network such as the Internet. A drive 1710 is also connected to the I / O interface 1705 as needed. A removable medium 1711, such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc., is installed on the drive 1710 as needed, so that a computer program read therefrom is installed into the storage section 1708 as needed.
[0202] In particular, according to an embodiment of the present disclosure, the process described above with reference to the flowchart can be implemented as a computer software program. For example, an embodiment of the present disclosure includes a computer program product, which includes a computer program carried on a computer-readable storage medium, and the computer program includes a program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from a network through a communication part 1709, and / or installed from a removable medium 1711. When the computer program is executed by a central processing unit (CPU) 1701, the above-mentioned functions defined in the system of the present application are executed.
[0203] It should be noted that the computer-readable storage medium shown in the present application may be a computer-readable signal medium or a computer-readable storage medium or any combination of the above two. The computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present application, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in combination with an instruction execution system, device or device. In the present application, a computer-readable signal medium may include a data signal propagated in a baseband or as part of a carrier wave, in which a computer-readable program code is carried. This propagated data signal may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. The computer-readable signal medium may also be any computer-readable storage medium other than a computer-readable storage medium that can send, propagate or transmit a program for use by or in conjunction with an instruction execution system, apparatus or device. The program code contained on the computer-readable storage medium may be transmitted using any appropriate medium, including but not limited to: wireless, wire, optical cable, RF, etc., or any suitable combination of the above.
[0204] The flow chart and block diagram in the accompanying drawings illustrate the possible architecture, function and operation of the system, method and computer program product according to various embodiments of the present application. In this regard, each box in the flow chart or block diagram can represent a module, a program segment or a part of a code, and the above-mentioned module, program segment or a part of a code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram or flow chart, and the combination of the boxes in the block diagram or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.
[0205] The units involved in the embodiments described in the present application may be implemented by software or hardware. The units described may also be set in a processor, for example, it may be described as: a processor includes a sending unit, an acquiring unit, a determining unit, and a first processing unit. The names of these units do not constitute limitations on the units themselves in some cases.
[0206] As another aspect, the present application also provides a computer-readable storage medium, which may be included in the device described in the above embodiment; or it may exist independently without being assembled into the device. The above computer-readable storage medium carries one or more programs, and when the above one or more programs are executed by a device, the device can implement functions including: obtaining target product features of a target product; obtaining object features of a target object; processing the target product features and the object features through a first machine learning model to obtain a first processing result; determining a first target object of the target product based on the first processing result to recommend the target product to the first target object.
[0207] Through the description of the above implementation methods, it is easy for those skilled in the art to understand that the example implementation methods described here can be implemented by software, or by combining software with necessary hardware. Therefore, the technical solution of the embodiment of the present disclosure can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, a USB flash drive, a mobile hard disk, etc.), including a number of instructions for a computing device (which can be a personal computer, a server, a mobile terminal, or a smart device, etc.) to execute the method according to the embodiment of the present disclosure, for example Figure 3 One or more of the steps shown.
[0208] In addition, the above-mentioned figures are only schematic illustrations of the processes included in the method according to the exemplary embodiments of the present disclosure, and are not intended to be limiting. It is easy to understand that the processes shown in the above-mentioned figures do not indicate or limit the time sequence of these processes. In addition, it is also easy to understand that these processes can be performed synchronously or asynchronously, for example, in multiple modules.
[0209] Those skilled in the art will readily appreciate other embodiments of the present disclosure after considering the specification and practicing the disclosure disclosed herein. The present disclosure is intended to cover any variations, uses, or adaptations of the present disclosure that follow the general principles of the present disclosure and include common knowledge or customary technical means in the art that are not applied for by the present disclosure. The specification and embodiments are to be regarded as exemplary only, and the true scope and spirit of the present disclosure are indicated by the claims.
[0210] It should be understood that the present disclosure is not limited to the detailed structures, drawings or implementations shown herein, but rather the present disclosure is intended to cover various modifications and equivalent arrangements included within the spirit and scope of the appended claims.
Claims
1. A product recommendation method, characterized in that: include: Obtaining target product features of a target product, wherein each target product is described by at least one system, each system includes at least one subsystem, each subsystem includes at least one option, and one system represents a description dimension; Obtain object features of a target object; Processing the target product features and the object features through a first machine learning model to obtain a first processing result; Determine a first target object of the target product according to the first processing result, so as to recommend the target product to the first target object; Wherein, obtaining the target product characteristics of the target product includes: The describer selects at least one option in each subsystem as a selected option to evaluate the target product based on his / her own experience or according to a scoring rule, wherein the describer is a person who has used the target product or is familiar with the target product; generating a vector of at least one dimension according to the selected options in the subsystem; Concatenate at least one-dimensional vectors of each subsystem in each system to generate a feature vector corresponding to each system; Obtaining N characteristic vectors corresponding to N systems of the target product, where N is a positive integer greater than or equal to 1; The target product features are determined according to the N feature vectors.
2. The method according to claim 1, characterized in that Get the object characteristics of the target object, including: Acquire historical product features of historical products, wherein the historical products refer to products of the same or similar product type as the target product, and the historical products are products that have been publicly released or sold or used by users; The object characteristics are determined based on the historical product characteristics.
3. The method according to claim 2, characterized in that Determining the object feature according to the historical product feature includes: Obtaining a historical duration of use of the historical product by the target object within a first predetermined time; Determine the weight of the corresponding historical product feature according to the historical duration; The object features are obtained according to the historical product features and their weights.
4. The method according to claim 2, characterized in that: Also includes: Get the first training set; The first machine learning model is trained using the first training set.
5. The method according to claim 4, characterized in that Get the first training set, including: Randomly identify a training product from historical products, and obtain training product features of the training product; randomly determining a training object from the target object, and obtaining a training object feature of the training object; If the training object has used the training product within the second predetermined time, generating positive samples in the first training set according to the training product characteristics and the training object characteristics; If the training object has not used the training product within the second predetermined time, a negative sample in the first training set is generated according to the training product characteristics and the training object characteristics.
6. The method according to claim 2, characterized in that Get historical product features of historical products, including: Obtain N feature vectors of the historical product, where N is a positive integer greater than or equal to 1; The historical product features are determined according to the N feature vectors.
7. The method according to any one of claims 1 to 6, characterized in that Also includes: Processing the object features by a second machine learning model to obtain a second processing result; A second target object of the target product is determined according to the second processing result, so as to recommend the target product to the second target object.
8. A product recommendation device, characterized in that: include: A product feature acquisition module configured to acquire target product features of a target product, wherein each target product is described by at least one system, each system includes at least one subsystem, each subsystem includes at least one option, and one system represents a description dimension; An object feature acquisition module configured to acquire object features of a target object; A first result acquisition module is configured to process the target product feature and the object feature through a first machine learning model to obtain a first processing result; A first target determination module, configured to determine a first target target of the target product according to the first processing result, so as to recommend the target product to the first target target; Wherein, obtaining the target product characteristics of the target product includes: The describer selects at least one option in each subsystem as a selected option to evaluate the target product based on his / her own experience or according to a scoring rule, wherein the describer is a person who has used the target product or is familiar with the target product; generating a vector of at least one dimension according to the selected options in the subsystem; Concatenate at least one-dimensional vectors of each subsystem in each system to generate a feature vector corresponding to each system; Obtaining N characteristic vectors corresponding to N systems of the target product, where N is a positive integer greater than or equal to 1; The target product features are determined according to the N feature vectors.
9. An electronic device, characterized in that: include: one or more processors; A storage device for storing one or more programs, when the one or more programs are executed by the one or more processors, enables the one or more processors to implement the product recommendation method as described in any one of claims 1 to 7.
10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the product recommendation method according to any one of claims 1 to 7 is implemented.
