Information recommendation method and device, computer device, and storage medium

Abstract

The application relates to the technical field of Internet application, and provides an information recommendation method and device, equipment and a storage medium, which comprises the following steps: obtaining a plurality of application objects pre-classified in an application program; calculating first upper limit confidence interval values of the application objects according to first historical operation information of a first user related to each application object; determining at least one second user similar to the first user according to portrait information of the first user; calculating second upper limit confidence interval values of each application object corresponding to each second user according to second historical operation information of the second user related to each application object; determining target upper limit confidence interval values of the application objects according to the first upper limit confidence interval values and the second upper limit confidence interval values of the application objects; and recommending each application object to the first user after sorting the application objects according to the target upper limit confidence interval values of the application objects. The method can realize accurate information recommendation and recommendation of unpopular information and newly listed information.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of Internet application, and particularly relates to an information recommendation method and device, computer equipment and a storage medium. BACKGROUND

[0002] With the development of the Internet and big data technology, information recommendation for users has always been a very popular and important research content. Although the current information recommendation algorithm is relatively mature, it also has some drawbacks.

[0003] For example, the collaborative recommendation method based on items or users can achieve relatively accurate recommendation results, but the calculation is complex and the recommendation effect for unpopular items is poor; for another example, the recommendation method based on association rules is simple and easy to operate, but the recommendation effect is poor.

[0004] Therefore, how to find a recommendation method that can guarantee the recommendation effect and push unpopular items has become a problem to be solved. SUMMARY

[0005] Therefore, it is necessary to provide an information recommendation method, device, computer equipment and storage medium which can achieve accurate pushing and push unpopular information in order to solve the above technical problems.

[0006] The embodiment of the present application provides an information recommendation method, which comprises the following steps:

[0007] Obtaining a plurality of application objects pre-classified in an application program, calculating a first upper limit confidence interval value of each application object according to first historical operation information related to the application object of a first user;

[0008] Determining at least one second user similar to the first user according to portrait information of the first user;

[0009] Calculating a second upper limit confidence interval value of each application object corresponding to each second user according to second historical operation information related to the application object of the second user;

[0010] Determining a target upper limit confidence interval value of each application object according to the first upper limit confidence interval value of each application object and the second upper limit confidence interval value of each application object corresponding to each second user;

[0011] Ranking each application object according to the target upper limit confidence interval value of each application object and recommending the application object to the first user.

[0012] In an embodiment, the determining the target upper confidence bound value of each application object according to the first upper confidence bound value of each application object and the second upper confidence bound value of each application object comprises:

[0013] Taking any one of the application objects as a first application object;

[0014] Calculating an average upper confidence bound value of the first application object according to the second upper confidence bound value of the first application object corresponding to each second user;

[0015] Determining a target upper confidence bound value of the first application object according to the first upper confidence bound value of the first application object and the average upper confidence bound value of the first application object.

[0016] In an embodiment, the determining the target upper confidence bound value of the first application object according to the first upper confidence bound value of the first application object and the average upper confidence bound value of the first application object comprises:

[0017] Obtaining a first weight corresponding to the first upper confidence bound value of the first application object according to a preset;

[0018] Determining a second weight corresponding to the average upper confidence bound value of the first application object according to the first weight; the sum of the first weight and the second weight is one;

[0019] Multiplying the first weight and the first upper confidence bound value of the first application object to obtain a first weighted value of the first application object;

[0020] Multiplying the second weight and the average upper confidence bound value of the first application object to obtain a second weighted value of the first application object;

[0021] Calculating the sum of the first weighted value and the second weighted value to obtain the target upper confidence bound value of the first application object.

[0022] In an embodiment, the

[0023] The obtaining the first weight corresponding to the first upper confidence bound value of the first application object according to a preset comprises:

[0024] Obtaining a plurality of weight functions stored locally;

[0025] Taking a weight function satisfying a preset condition as a target weight function;

[0026] Determining the first weight of the first application object according to the target weight function.

[0027] In an embodiment, the weight function satisfying the preset condition is taken as the target weight function, comprising:

[0028] Based on the preset evaluation function, an evaluation function value corresponding to each weight function is determined respectively;

[0029] The weight function with the largest evaluation function value is determined as the target weight function.

[0030] In an embodiment, the application object includes a commodity, and the first upper limit confidence interval value of each application object is calculated according to the first historical operation information related to each application object, comprising:

[0031] According to the first historical operation information, the order placing times of the first user for each commodity are determined;

[0032] The historical recommendation data of the first user recommending each commodity stored locally is obtained;

[0033] According to the historical recommendation data, the total times of recommending all commodities to the first user and the recommendation times of recommending each commodity to the first user are determined;

[0034] According to the order placing times of the first user for each commodity, the total times of recommending all commodities to the first user, and the recommendation times of recommending each commodity to the first user, the first upper limit confidence interval value of each commodity is determined;

[0035] Wherein, the first upper limit confidence interval value of each commodity is positively correlated with the order placing times of the first user for each commodity and the total times of recommending all commodities to the first user, and is negatively correlated with the recommendation times of recommending each commodity to the first user.

[0036] In an embodiment, the method further comprises:

[0037] When the recommendation times of recommending each commodity to the first user is 0, the first upper limit confidence interval value of the commodity with the recommendation times of 0 is 0.

[0038] Embodiments of the present application provide an information recommendation device, comprising:

[0039] An acquisition module is configured to acquire a plurality of application objects pre-classified in an application program;

[0040] A first calculation module is configured to calculate a first upper limit confidence interval value of each application object according to first historical operation information related to each application object of a first user;

[0041] a matching module configured to determine at least one second user similar to the first user according to profile information of the first user;

[0042] a second calculating module configured to calculate a second upper confidence interval value of each application object corresponding to each second user according to second historical operation information of each application object related to the second user;

[0043] a determining module configured to determine a target upper confidence interval value of each application object according to the first upper confidence interval value of each application object and the second upper confidence interval value of each application object corresponding to each second user;

[0044] a recommending module configured to recommend each application object to the first user after sorting each application object according to the target upper confidence interval value of each application object.

[0045] In one embodiment, the determining module comprises:

[0046] a first determining unit configured to take any application object in the application objects as a first application object;

[0047] a calculating unit configured to calculate an average upper confidence interval value of the first application object according to the second upper confidence interval value of the first application object corresponding to each second user;

[0048] a second determining unit configured to determine a target upper confidence interval value of the first application object according to the first upper confidence interval value of the first application object and the average upper confidence interval value of the first application object.

[0049] In one embodiment, the second determining unit is further configured to:

[0050] obtain a first weight corresponding to the first upper confidence interval value of the first application object;

[0051] determine a second weight corresponding to the average upper confidence interval value of the first application object according to the first weight; the sum of the first weight and the second weight is one;

[0052] multiply the first weight and the first upper confidence interval value of the first application object to obtain a first weighted value of the first application object;

[0053] multiply the second weight and the average upper confidence interval value of the first application object to obtain a second weighted value of the first application object;

[0054] calculate the sum of the first weighted value and the second weighted value to obtain the target upper confidence interval value of the first application object.

[0055] In an embodiment, the second determining unit is further configured to:

[0056] obtain a plurality of weight functions stored locally;

[0057] determine a target weight function that satisfies a preset condition;

[0058] determine the first weight of the first application object according to the target weight function.

[0059] In an embodiment, the second determining unit is further configured to:

[0060] determine an evaluation function value corresponding to each of the weight functions based on a preset evaluation function;

[0061] determine the target weight function as the weight function with the largest evaluation function value.

[0062] In an embodiment, the application object includes a commodity, and the first calculating module is further configured to:

[0063] determine the order placing frequency of each of the commodities by the first user according to the first historical operation information;

[0064] obtain historical recommendation data for recommending each of the commodities to the first user stored locally;

[0065] determine the total number of times of recommending all commodities to the first user and the number of times of recommending each of the commodities to the first user according to the historical recommendation data;

[0066] determine the first upper limit confidence interval value of each of the commodities according to the order placing frequency of each of the commodities by the first user, the total number of times of recommending all commodities to the first user, and the number of times of recommending each of the commodities to the first user;

[0067] wherein the first upper limit confidence interval value of each of the commodities is positively correlated with the order placing frequency of each of the commodities by the first user and the total number of times of recommending all commodities to the first user, and is negatively correlated with the number of times of recommending each of the commodities to the first user.

[0068] In an embodiment, the first calculating module is further configured to:

[0069] when the number of times of recommending each of the commodities to the first user is 0, determine that the first upper limit confidence interval value of the commodity with the number of times of recommendation of 0 is 0.

[0070] The embodiment of the present application provides a computer device, comprising a memory and a processor, the memory stores a computer program, and the processor implements the steps of the information recommendation method provided by any embodiment of the present application when executing the computer program.

[0071] The embodiment of the present application provides a computer readable storage medium, which stores a computer program, and the computer program is characterized in that, when the processor executes the computer program, the steps of the information recommendation method provided by any embodiment of the present application are implemented.

[0072] The information recommendation method, device, computer device and storage medium provided by the embodiment of the present application first acquire a plurality of application objects pre-classified in an application program, calculate first upper limit confidence interval values of the application objects according to first historical operation information related to the application objects of a first user, determine at least one second user similar to the first user according to portrait information of the first user, and calculate second upper limit confidence interval values of the application objects corresponding to each second user according to second historical operation information related to the application objects of the second user, then determine target upper limit confidence interval values of the application objects according to the first upper limit confidence interval values of the application objects and the second upper limit confidence interval values of the application objects corresponding to each second user, and further sort the application objects according to the target upper limit confidence interval values of the application objects and recommend the application objects to the first user. Since the target upper limit confidence interval values of the application objects are determined according to the upper limit confidence interval values of the same application object of the first user and similar users, when information is recommended, the preferences of the user and the preferences of similar users are considered, the accuracy of information recommendation is ensured, and the recommendation of unpopular information and newly listed information can be realized. BRIEF DESCRIPTION OF DRAWINGS

[0073] Figure 1 An application scenario diagram of the information recommendation method in one embodiment is shown in the figure;

[0074] Figure 2 A flowchart of the information recommendation method in one embodiment is shown in the figure;

[0075] Figure 3 A flowchart of the information recommendation method in another embodiment is shown in the figure;

[0076] Figure 4 A flowchart of the product recommendation method provided by the embodiment of the present application is shown in the figure;

[0077] Figure 5 A structural block diagram of the information recommendation device in one embodiment is shown in the figure;

[0078] Figure 6 An internal structure diagram of the computer device in one embodiment is shown in the figure. DETAILED DESCRIPTION

[0079] In order to make the purpose, technical scheme and advantages of the present application more clear, the present application is further described in detail below in combination with the drawings and examples. It should be understood that the specific examples described herein are only used to explain the present application and do not limit the present application.

[0080] In the e-commerce industry, product recommendation for users is a popular research content. The commonly used product recommendation methods and their corresponding advantages and disadvantages are:

[0081] (1) Recommendation method based on product content. For example, food A and food B, for their price, taste, shelf life, brand and other dimensions, the similarity can be calculated, and the product recommendation is made according to the similarity. The pre-processing of this recommendation method is complex, the product dimension is more, it is difficult to select the appropriate dimension for similarity calculation, which leads to poor recommendation effect.

[0082] (2) Recommendation method based on association rules. The most common one is to recommend products based on user's buying habits, such as the classic "beer diapers" case. However, in actual operation, this method is rarely used because it requires sufficient data for association rules and the recommendation accuracy is poor.

[0083] (3) Collaborative recommendation method based on items. Suppose item A is purchased by Xiaozhang, Xiaoming and Xiaodong, item B is purchased by Xiaohong, Xiali and Xiaochen, and item C is purchased by Xiaozhang, Xiaoming and Xiaoli. At first glance, the purchase crowd similarity of item A and item C is higher (relative to item B), so item C can be recommended to Xiaodong and item A can be recommended to Xiaoli. This method can achieve good recommendation effect, but the calculation is complex and the recommendation effect for unpopular items is poor.

[0084] (4) Recommendation method based on deep learning. For example, convolutional neural network, recurrent neural network, deep neural network and other methods are used in product recommendation. This recommendation method can achieve good recommendation effect, but the application is not mature at present, and the parameter adjustment during model training is difficult.

[0085] To solve the above problems, the information recommendation method provided in the application is a brand-new recommendation algorithm found based on the Monte Carlo strategy. The core idea of the Monte Carlo tree search strategy is to calculate the Upper Confidence Bound (UCB) value and select the subsequent node with the maximum UCB value. The inventors found in the research process that, for product recommendation, the large classification, secondary classification, fine classification, store, product model, etc. of the product are equivalent to the nodes of the Monte Carlo tree, and therefore the inventors thought of applying the Monte Carlo tree search strategy to product recommendation, and even to information recommendation such as information recommendation and music recommendation. By using the scheme of the application, on the one hand, the user can be accurately pushed to ensure that the user receives the most favorite and most commonly used information, and on the other hand, the latest information can be pushed to avoid missing unpopular information and newly listed information, and to avoid the predicament of new users being unable to be recommended information due to no historical data.

[0086] The information recommendation method provided in the application can be applied to an application environment as shown in Figure 1 , and specifically applied to the server 104 in Figure 1 . Figure 1 In the application, the server 104 can be implemented by an independent server or a server cluster composed of multiple servers, and can be a server corresponding to an application program with information push requirements such as an e-commerce platform, a short video platform, a music platform, and a news information platform. The above-mentioned application program can be installed in the electronic device 102, and the electronic device 102 can be a computer, a mobile phone, a wearable device, etc. In the embodiment of the application, the electronic device 102 and the server 104 communicate through a network. Taking product recommendation as an example, when the application is applied to product recommendation of an e-commerce platform, the application objects include products or categories of products. After a first user logs in to the e-commerce platform through the electronic device 102, the server 104 corresponding to the e-commerce platform obtains a plurality of application objects pre-classified in the e-commerce platform, calculates the first UCB value of each application object according to the first historical operation information of the first user related to each application object, determines at least one second user similar to the first user according to the portrait information of the first user, calculates the second UCB value of each application object corresponding to each second user according to the second historical operation information of the second user related to each application object, and then determines the target UCB value of each application object according to the first UCB value of each application object and the second UCB value of each application object corresponding to each second user, sorts each application object according to the target UCB value of each application object, and displays the sorting result in the e-commerce platform of the electronic device 102 to recommend to the first user, thereby realizing accurate push of products and pushing of unpopular and newly listed products.

[0087] In one embodiment, as shown in Figure 2As shown, an information recommendation method is provided. The embodiment mainly illustrates the information recommendation method applied to a server in an application. Figure 1 As shown in the figure, the information recommendation method can include the following steps: Figure 2

[0088] Step 202, obtaining a plurality of application objects pre-classified in the application program, and calculating a first upper confidence interval value of each application object according to first historical operation information of the first user related to each application object.

[0089] application program refers to any application with information recommendation function, including but not limited to e-commerce platform, short video platform, music platform, news information platform, etc. Application object refers to the recommended object corresponding to the application program, and the application object is pre-classified and determined. Different application programs correspond to different application objects. For example, for an e-commerce platform, the pre-classified application objects can include goods, categories to which the goods belong (such as home appliances, 3C digital, clothing, food, etc.), brands of goods, etc. For a music platform, the pre-classified application objects can include music, music categories to which the music belongs (such as folk, pop, rock, etc.). For a news information platform, the pre-classified application objects can include news, news categories to which the news belongs (such as entertainment, finance, automobile, Internet, etc.). The first user refers to any user currently using an application program with information recommendation function, and the first historical operation information refers to the historical operation record of the first user on the currently used application program, including but not limited to browsing record, order record, collection record, sharing record, etc.

[0090] It can be understood that in the embodiment of the application, for different application programs, the obtained first historical operation information can be different. For example, for an e-commerce platform, the obtained first historical operation information can include the browsing record, order record, add-to-cart record, and collection record of the first user on the e-commerce platform. For example, for a short video platform, the obtained first historical operation information can include the record of the first user watching videos, liking videos, sharing videos, commenting on videos, and collecting videos. The first historical operation information contains the operation record of the first user on different application objects. As the first user uses the application program for a longer time, the application objects contained in the first historical operation information will increase.

[0091] In the embodiment of the application, when the server recommends information to the first user, the operation record of the first user on each application object in the application program can be obtained from the local memory or the database where the user's operation record is stored as the first historical operation information, and the first UCB value of each application object can be calculated according to the obtained first historical operation information.

[0092] ​It can be understood that the number of first UCB values is consistent with the number of application objects, and several first UCB values are calculated for several application objects, and one application object corresponds to one first UCB value.

[0093] Exemplarily, taking music recommendation as an example, the application objects include music and categories to which the music belongs, the server can analyze the total duration of the first user listening to all songs and the duration of the first user listening to each song from the first historical operation record, the duration of the first user listening to the songs of each category, determine the first UCB value of each song according to the duration of the first user listening to each song and the total duration of listening to all songs, and determine the first UCB value of each song category according to the duration of the first user listening to the songs of each category and the total duration of listening to all songs.

[0094] Exemplarily, taking music recommendation as an example, the application objects include music and categories to which the music belongs, the server can analyze the total duration of the first user listening to all songs and the duration of the first user listening to each song from the first historical operation record, the duration of the first user listening to the songs of each category, determine the first UCB value of each song according to the duration of the first user listening to each song and the total duration of listening to all songs, and determine the first UCB value of each song category according to the duration of the first user listening to the songs of each category and the total duration of listening to all songs.

[0095]

[0096] Among them, represents the average reward in the past, which can be defined differently. N represents the total number of times that the server recommends all information (such as songs, goods, etc.) to the first user, n j represents the number of times that the server recommends the jth application object, and C is a preset constant, such as C can be set to 1, UCB userj represents the first UCB value of the jth application object for the first user. In music recommendation, which can be defined as the number of times of listening, or the duration of listening to songs, the number of sharing times, etc.; N represents the total number of times that the server recommends all songs to the first user, n j represents the number of times that the server recommends the jth song or song category to the first user. N and n j can be analyzed by the server according to the historical recommendation data of the server for song recommendation to the first user, or the latest N and n j can be stored by the server, and after each music recommendation, the number of songs recommended this time and the songs recommended this time and the song categories to which they belong are counted, and N and n j are updated according to the number of songs recommended this time and the songs recommended this time and the song categories to which they belong.

[0097] It should be noted that, as shown in formula (1) above, when the j-th application object has never been recommended by the server, the UCB value cannot be calculated using formula (1). In this case, the UCB value can be preset to 0. That is to say, when n j When n is 0, the server determines that the first user's first UCB value for the j-th application object is 0. Therefore, regardless of n... j Whether the value is 0 or not, the UCB value of the first user for each information object can be obtained.

[0098] Step 204: Determine at least one second user similar to the first user based on the first user's profile information.

[0099] In this embodiment of the application, the server can, with the authorization of the first user, obtain the registration information of the first user on the application from the local storage space based on the login account of the first user, analyze the profile information of the first user based on the obtained registration information, and then determine at least one second user similar to the first user based on the profile information of the first user.

[0100] For example, based on the first user's registration information, the first user's profile information can be divided into several dimensions or even more dimensions as shown in Table 1.

[0101] Table 1

[0102] Name Zhang City Shanghai Gender Male Hometown Shandong Age 26 Education College Marital Status No Identity Staff Monthly Income 5k Housing Status None Hobbies Football

[0103] In this embodiment of the application, after obtaining the profile information of the first user, each dimension of the profile information can be regarded as a vector, and the profile information of the first user can be represented as S(v1,v2,v3…,v…). n ), where v1~v n Let n vectors represent the first user's profile information. Each vector represents one dimension of the first user's profile information. Then, based on the represented first user's profile information, the server can use big data clustering algorithms such as K-means clustering and K-centroid clustering to find at least one similar user belonging to the same group as the first user from all users of the application. This user is called the second user.

[0104] Step 206: Calculate the second upper limit confidence interval value for each application object corresponding to each second user based on the second historical operation information of the second user related to each application object.

[0105] The second historical operation information is operation records generated by each second user in the past when using the application program, and the content of the second historical operation information is similar to that of the first historical operation information. It can be understood that one second user corresponds to the second historical operation information of the second user itself. In order to solve the problem that the prior art cannot recommend goods that the user has not browsed or purchased, newly listed goods, unpopular goods, etc., in the embodiment of the application, similar second users to the first user are found, for example, 100 people, and the UCB value of each second user for each application object is calculated, which is called a second UCB value.

[0106] In the embodiment of the application, after obtaining the second historical operation information of each second user, the second UCB value of each second user for each application object can be calculated according to the second historical operation information of each second user.

[0107] It can be understood that for the second user A, the second UCB value of the second user A for each application object is calculated by using the second historical operation information of the second user A; for the second user B, the second UCB value of the second user B for each application object is calculated by using the second historical operation information of the second user B. There are several application objects, and each second user corresponds to several second UCB values.

[0108] It should be noted that the calculation method of the second UCB value is similar to that of the first UCB value, which will not be described here.

[0109] Step 208, determining a target upper confidence interval value of each application object according to the first upper confidence interval value of each application object and the second upper confidence interval value of each application object corresponding to each second user.

[0110] In the embodiment of the application, after the first UCB value and the second UCB value of each application object are respectively calculated, the target UCB value corresponding to each application object can be determined according to the first UCB value of each application object and the second UCB value of each application object.

[0111] For example, for the jth application object, the first UCB value of the first user for the jth application object and the second UCB value of each second user for the jth application object are summed, and then the sum is divided by the total number of second users and first users to obtain a mean value, and the mean value is taken as the target UCB value corresponding to the jth application object.

[0112] Step 210, ranking each application object according to the target upper confidence interval value of each application object and recommending the first user.

[0113] In the embodiments of the present application, after the target UCB value corresponding to each application object is determined, the application objects can be sorted according to the size of the target UCB value of each application object, and the sorted application objects are recommended to the first user.

[0114] Exemplarily, the application objects can be sorted in descending order of the target UCB value, the application objects with larger target UCB value are arranged in front, and the application objects with smaller target UCB value are arranged in back, and the server pushes the sorted application objects to the application program for display, so that the sorted application objects are displayed to the first user through the application program.

[0115] It should be noted that when the server sorts the application objects according to the target UCB value, the application objects of the same level can be sorted separately, for example, when recommending goods, the goods of the same level can be sorted separately according to the target UCB value of the goods of the same level, the subcategories can be sorted separately according to the target UCB value of the subcategories, and the goods of the same category can be sorted separately according to the target UCB value corresponding to different brands. Alternatively, the goods and different levels of categories can be mixed and sorted together. The sorting mode can be set according to the information recommendation requirements of the application program, and the present application does not limit this.

[0116] The information recommendation method provided in the embodiments of the present application first acquires a plurality of application objects pre-classified in the application program, calculates the first upper confidence interval value of each application object according to the first historical operation information of the first user related to each application object, determines at least one second user similar to the first user according to the portrait information of the first user, and calculates the second upper confidence interval value of each application object corresponding to each second user according to the second historical operation information of the second user related to each application object. Then, the target upper confidence interval value of each application object is determined according to the first upper confidence interval value of each application object and the second upper confidence interval value of each application object corresponding to each second user, and each application object is sorted and recommended to the first user according to the target upper confidence interval value of each application object. Since the target UCB value corresponding to each application object is determined according to the UCB value of the same application object by the first user and the similar second user, the preferences of the similar users are combined when the information is recommended, and the recommendation of the cold information and the newly listed information can be realized while ensuring the accuracy of the information recommendation.

[0117] In an optional embodiment of the present application, as shown in Figure 3 based on the embodiment shown in Figure 2 Step 208 can include the following steps:

[0118] Step 302, any application object in the application objects is taken as a first application object.

[0119] In step 304, the average upper confidence bound value of the first application object is calculated according to the second upper confidence bound value of the first application object corresponding to each second user.

[0120] In the embodiments of the present application, any application object in the application objects can be taken as the first application object, and the determination manner of the target UCB value of each application object is explained by taking the determination manner of the target UCB value of the first application object as an example. Specifically, after the second UCB value of each application object corresponding to each second user is calculated, the UCB mean value of the first application object is calculated according to the second UCB value of the first application object corresponding to each second user, and the average UCB value of the first application object is obtained.

[0121] In step 306, the target upper confidence bound value of the first application object is determined according to the first upper confidence bound value of the first application object and the average upper confidence bound value of the first application object.

[0122] In an optional embodiment of the present application, the sum of the first UCB value of the first application object and the average UCB value of the first application object is calculated, and the obtained sum value is taken as the target UCB value of the first application object.

[0123] In an optional embodiment of the present application, a preset first weight corresponding to the first upper confidence bound value of the first application object is obtained; a second weight corresponding to the average upper confidence bound value of the first application object is determined according to the first weight, wherein the sum of the first weight and the second weight is one; the first weight is multiplied by the first upper confidence bound value of the first application object to obtain a first weighted value of the first application object, and the second weight is multiplied by the average upper confidence bound value of the first application object to obtain a second weighted value of the first application object; and then the sum of the first weighted value and the second weighted value is calculated to obtain the target upper confidence bound value of the first application object.

[0124] The first weight corresponding to the first UCB value of the first application object can be a fixed parameter set in advance, and the value is any value in the interval of 0-1, for example, the first weight can be set to 0.5 to balance the data of the first user and the data of the similar second user. Alternatively, considering that the user's historical data is small when the user first uses the application, the reference value is small, and the first weight can take a small value. With the user's use, the historical data is more and more, and the value of the first weight can gradually increase, but considering that it is also hoped to recommend new information, unpopular information, etc. to the user, the value of the first weight can not exceed 0.5 to avoid that new information, unpopular information, and information that has not been recommended cannot be recommended. Therefore, the first weight can also be dynamically determined according to a preset weight function, and the variable of the weight function can be time (days, months, years, etc.), order quantity, historical record, user characteristics, etc.

[0125] Exemplarily, when the first weight is determined according to the preset weight function, the variable of the weight function is the number of months that the user uses the application, the number of months that the user currently uses the application can be used to determine the first weight, for example, the first user has used the application for three months, and the three months are brought into the weight function, and the weight function value obtained is used as the first weight.

[0126] Then, after the first weight is determined, the difference between 1 and the first weight can be used as the second weight corresponding to the average UCB value of the first application object.

[0127] Further, the product of the first weight and the first UCB value of the first application object is calculated to obtain a first weighted value, and the product of the second weight and the average UCB value of the first application object is calculated to obtain a second weighted value, and the sum of the first weighted value and the second weighted value is calculated to obtain the target UCB value of the first application object.

[0128] Specifically, the target UCB value of the first application object can be determined by the following formula (2):

[0129] UCB total target = a*UCB user first + (1-a)*UCB average average (2)

[0130] UCB total target represents the target UCB value of the first application object, UCB user first represents the first UCB value of the first application object, UCB average average represents the average UCB value of the first application object, and a represents the first weight.

[0131] In the embodiments of the present application, when the first weight a is determined according to a preset weight function, that is, the first weight a is defined as a dynamic weight function, the weight function can be a preset function meeting some conditions, wherein the conditions met by the weight function are as follows:

[0132] (1) a needs to be in the interval of 0-1, preferably, a is in the interval of 0-0.5;

[0133] (2) the weight function needs to change with time, the number of times of recommendation of a user, the use condition of a user, preference, etc.;

[0134] (3) the weight function has good convergence, and can approach a limit value, such as 0.5, with time, historical data and use condition;

[0135] (4) the variable of the weight function can be time, the number of times of recommendation, the order amount, the type of browsed goods, etc., but a value needs to approach a limit value, such as 0.5, after a period of time (such as one year), that is, the UCB value of any application object includes 50% of the UCB value and 50% of the UCB value. user average .

[0136] In the embodiments of the present application, a can be defined based on any function in functions meeting the above conditions, such as a Sigmod function, a hyperbolic tangent Tanh function and a Relu function.

[0137] For example, a = 0.5 * S(x) is defined, wherein S(x) represents a Sigmod function, x can be a time variable such as a day, a week or a month, or can be an order amount or the number of times of order, and can be selected according to convergence.

[0138] For example, a = 0.5 * tanh(x) is defined, wherein x can be a time variable such as a day, a week or a month, or can be an order amount or the number of times of order, and can be selected according to convergence.

[0139] In the embodiments of the present application, a is defined as a weight function meeting some conditions, and is limited with time, the use condition of a user, preference and the success rate of recommendation, and the main reason is that when a user starts to use, the application program does not recommend the user or the amount of recommended data is small, and the n j value of most application objects is 0, that is, the UCB user value is 0, and there is no UCB user initial value, and the data UCB average ​The application program recommends the goods to the user when the user uses the application program for a certain period of time. The goods that the user has ordered, or added to the shopping cart, and which the user is interested in and needs to browse multiple times to decide whether to order, are recommended more times, that is, UCB user The goods with high values are recommended preferentially to improve the purchase desire of the user.

[0140] By adjusting a, the user is recommended familiar and interesting data goods on one hand, and is recommended new goods, goods that have not been browsed and recommended, that is, UCB on the other hand average The goods with high values; by adjusting a, the user is recommended goods that the user is interested in, and the quantity of goods ordered by the user is increased, that is, UCB total The value is increased to achieve the purpose of accurate pushing.

[0141] In an optional embodiment of the present application, the best and most suitable weight function can be selected from multiple weight functions that meet the conditions to define a to achieve the best recommendation effect. Specifically, multiple weight functions stored locally can be obtained, and the weight function that meets the preset condition is used as a target weight function, and then the first weight of the first application object is determined according to the target weight function.

[0142] The multiple weight functions stored locally can be weight functions that meet the conditions (1)-(4) above, such as weight functions 0.5*Sigmod function, 0.5*Tanh function, and 0.5*Relu function defined based on Sigmod function, hyperbolic tangent Tanh function, and Relu function.

[0143] In an optional embodiment of the present application, the preset condition can be the highest revenue, that is, the weight function that obtains the highest revenue when the target UCB values of the application objects generated based on the first weights defined by the weight functions are used for recommendation is determined as the target weight function.

[0144] The revenue brought by the weight function can be evaluated by a preset evaluation function. Specifically, based on the preset evaluation function, an evaluation function value corresponding to each weight function is determined; the weight function with the largest evaluation function value is determined as the target weight function. Then, the first weight can be determined based on the determined target weight function.

[0145] For example, the preset evaluation function can be represented by the following formula (3):

[0146]

[0147] X represents the return of the first user each time using the recommendation, which can be defined differently according to different application programs, for example, in the commodity recommendation, X can be represented as the order number, order amount, etc. of the first user, in the news information recommendation, X can be represented as the browsing number, interested number, etc., and n represents the total number of recommendations of the server.

[0148] Exemplarily, in the commodity recommendation, taking the order number as an example, assuming that the total number of recommendations n is 1000, the evaluation function values of a defined as different weight functions or values are shown in Table 2.

[0149] Table 2

[0150]

[0151] It can be understood that the higher the value is, the better the recommendation effect is. As can be seen from Table 2, the recommendation effect of a defined as 0.5*Tanh function is the best, so it can be determined that the target weight function is 0.5*Tanh function, and then the first weight is determined based on 0.5*Tanh function.

[0152] It can be understood that the same weight function can obtain different benefits in different user groups with similar portraits, for example, the weight function defined based on sigmoid function can obtain the highest benefit in user group A, and can obtain lower benefit in user group B. Therefore, in the embodiments of the present application, in order to obtain the best recommendation effect, test users of different user groups can be selected for testing, wherein the users included in a user group have similar portraits, and the same user group can be further divided into multiple small groups, and each small group corresponds to a different weight function, that is, a small group corresponds to a recommendation strategy based on the weight function defined based on sigmoid function, a small group corresponds to a recommendation strategy based on the weight function defined based on tanh function, and a small group corresponds to a recommendation strategy based on the weight function defined based on Relu function. Based on the historical operation records and historical recommendation records of the application program used by these small groups, the return of the test user each time using the recommendation and the total number of recommendations are determined, and then the evaluation function values of different weight functions in the same user group are calculated based on the above formula (3). Further, when the first weight is obtained, multiple weight functions are obtained, and the test user group similar to the first user is determined according to the portrait information of the first user, then the evaluation function value of each weight function in the similar test user group is determined, and the weight function with the largest evaluation function value is selected as the target weight function to determine the first weight.

[0153] In the embodiments of the present application, the evaluation function value corresponding to each weight function is determined based on a preset evaluation function, and then the function with the maximum evaluation function value is determined as the target weight function, so that the target weight function with the best recommendation effect can be selected from multiple weight functions to define a, so as to achieve the best recommendation effect.

[0154] The information recommendation method provided in the embodiments of the present application takes any application object in each application object as a first application object, calculates the average UCB value of the first application object according to the second UCB value of the first application object corresponding to each second user, and then determines the target UCB value of the first application object according to the first UCB value of the first application object and the average UCB value of the first application object, so that the target UCB value of any application object calculated not only considers the user's own preference, but also considers the preference of similar users, which provides data support for subsequent information recommendation according to the UCB value, and is beneficial to improve the recommendation accuracy and realize the recommendation of unpopular information.

[0155] In an optional embodiment of the present application, the method can be applied to commodity recommendation of an e-commerce platform, so that the pre-classified application objects in the application program can include commodities and commodity categories, wherein the commodity categories can be large categories to which the commodities belong, such as home appliances, 3C digital products, clothing, food, etc., or can be sub-categories under each large category, such as mobile phones, notebook computers, tablet computers, earphones, etc. sub-categories under the 3C digital product category, or can be brands to which the commodities belong, stores where the commodities are located, etc. In the embodiments of the present application, when calculating the first UCB value of each application object according to the first historical operation information of the first user related to each application object, the number of orders of each commodity by the first user can be determined according to the first historical operation information, the historical recommendation data of recommending each commodity to the first user stored locally is obtained, and the total number of times of recommending all commodities to the first user and the number of times of recommending each commodity to the first user are determined according to the historical recommendation data, and then the first upper confidence interval value of each commodity is determined according to the number of orders of each commodity by the first user, the total number of times of recommending all commodities to the first user, and the number of times of recommending each commodity to the first user, wherein the first upper confidence interval value of each commodity is positively correlated with the number of orders of each commodity by the first user and the total number of times of recommending all commodities to the first user, and is negatively correlated with the number of times of recommending each commodity to the first user.

[0156] Exemplarily, the first UCB value of each commodity can be calculated by the above formula (1). When performing commodity recommendation, represents the number of orders of the jth commodity by the first user, N represents the total number of times of recommending all commodities to the first user by the server, and nj represents the number of times that the server recommends the jth commodity to the first user.

[0157] It can be understood that when the application object includes a commodity category, when determining the order placing times of each commodity category according to the first historical operation information, the commodity category to which each commodity in the first historical operation information belongs can be analyzed according to the first historical operation information, and then the order placing times of each commodity category are counted. Then, when calculating the first UCB value of the jth commodity category, represents the order placing times of a certain commodity category by the first user, N represents the total number of times that the server recommends all commodities to the first user, and n j represents the number of times that the server recommends the jth commodity category to the first user.

[0158] It should be noted that, represents the average reward in the past, which can be defined differently, such as It can be defined as the order placing times, the order placing amount, the number of times of interest (including adding to the shopping cart, browsing for more than 10 seconds, the order placing times, etc.), and the like. The order placing times are used in the present application as an example to explain the present application, and cannot be regarded as a limitation on the present application. Accordingly, N and n j It can also be defined differently, such as when N represents the total amount of all commodities recommended by the server to the first user, and n j represents the amount involved in the number of times that the server recommends the jth commodity.

[0159] It can be understood that the scheme of the present application can not only be applied to commodity recommendation, but also be applied to short video recommendation, music recommendation, news recommendation, and the like, N and n j The meaning of can be defined according to the specific application scene, which is not limited in the present application.

[0160] In addition, it should be noted that from the above formula (1), in commodity recommendation, when calculating the UCB value, it is necessary to ensure that n j is not 0 and is a positive value, that is, the jth commodity or the jth commodity category has been recommended by the server.

[0161] In an optional embodiment of the present application, when the jth commodity or the jth commodity category has never been recommended by the server, the UCB value cannot be calculated by the above formula (1). In this case, the UCB value can be set to 0 in advance. That is, when the recommendation times of each commodity to the first user are 0, the first upper limit confidence interval value of the commodity with the recommendation times of 0 is 0. Similarly, when the recommendation times of each commodity of the commodity category to the first user are 0, the first UCB value of the commodity category with the recommendation times of 0 is 0. Thus, regardless of whether n j is 0, the UCB value of each commodity or commodity category for the first user can be obtained.

[0162] The information recommendation method provided by the embodiments of the present application can be applied to scenarios such as commodity recommendation, music recommendation, and short video recommendation. Taking commodity recommendation as an example, Figure 4 The flowchart of the commodity recommendation method provided by an embodiment of the present application is shown in the figure.

[0163] As Figure 4 shown, the commodity recommendation method can include the following steps:

[0164] Step 401: Obtain a plurality of application objects pre-classified in an e-commerce platform, and calculate first UCB values of each application object according to first historical operation information related to each application object of a first user, wherein the application objects include commodities, commodity categories, commodity sub-categories, and brands of commodities.

[0165] Step 402: Determine at least one second user similar to the first user according to portrait information of the first user.

[0166] Step 403: Calculate second UCB values of each application object corresponding to each second user according to second historical operation information related to each application object of the second user.

[0167] Step 404: Calculate target UCB values of each application object according to the first UCB values of each application object and the second UCB values of each application object corresponding to each second user.

[0168] It should be noted that the description of steps 401-404 can refer to the related content of the foregoing embodiments, which will not be described here.

[0169] In the embodiments of the present application, target UCB values of each commodity, and each category, sub-category, and brand of the commodity are calculated respectively.

[0170] Step 405: Sort each commodity category according to the target UCB values of each commodity category and recommend the sorted commodity categories to the first user.

[0171] Step 406, according to the target UCB value of each commodity sub-category under the target commodity category selected by the first user, each commodity sub-category is sorted and recommended to the first user.

[0172] Step 407, according to the target UCB value of each brand under the target commodity sub-category selected by the first user, each brand is sorted and recommended to the first user.

[0173] Step 408, according to the target UCB value of each commodity under the target brand selected by the first user, each commodity is sorted and recommended to the first user.

[0174] Step 409, obtaining the operation record of the first user on each commodity.

[0175] The operation record includes but is not limited to browsing, collecting, adding to shopping cart, purchasing, etc.

[0176] Step 410, according to the operation record and the target commodity object, updating the target UCB value of the target commodity object of the first user.

[0177] The target commodity object includes target commodities, target commodity categories, target commodity sub-categories, target brands of target commodities, etc. browsed, clicked or purchased by the user.

[0178] In the embodiment of the application, each commodity category can be sorted and recommended to the user according to the target UCB value of each commodity category, such as commodity categories including home appliances, 3C electronic products, vegetables, home decoration, fruits, etc. After the user selects a commodity category, such as the user clicking into the 3C electronic product commodity category, each commodity sub-category under the 3C electronic product is sorted and recommended to the user according to the target UCB value of each commodity sub-category under the 3C electronic product, such as mobile phones, laptops, hair dryers, etc. After the user selects a commodity sub-category, such as the user clicking into the mobile phone commodity sub-category, each brand under the mobile phone category is sorted and recommended to the user according to the target UCB value of each brand under the mobile phone category, such as brand A, brand B, brand C, etc. After the user selects a brand, such as the user clicking into brand A, each commodity in brand A is sorted and recommended to the user according to the target UCB value of each commodity under brand A, and the operation record of the user on each commodity is obtained. Then, according to the operation record and the target brand, target commodity sub-category, target commodity category selected by the user, the target UCB value of the target commodity, target commodity category, target commodity sub-category, target brand of target commodity browsed, clicked or purchased by the current user is updated, so that the latest recommendation list is presented to the user when the user enters the e-commerce platform next time.

[0179] The scheme of the present application is applied to commodity recommendation, for the first time, the Monte Carlo search algorithm is used for commodity recommendation, on the one hand, accurate push can be carried out to the user, it is guaranteed that the user receives the favorite, most commonly used commodity information, on the other hand, the latest product information can be pushed, avoiding the omission of unpopular commodities, newly listed commodities, and the commodity preference data of similar groups can also be recommended, avoiding the plight that new users cannot be recommended due to no historical data.

[0180] It should be understood that, although Figure 2-4 The steps in the flowchart of the present application are displayed in sequence according to the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified herein, the execution of these steps is not strictly limited in sequence, and these steps can be executed in other orders. Moreover, Figure 2-4 At least part of the steps in the flowchart of the present application can include multiple sub-steps or multiple stages, which are not necessarily executed at the same time, but can be executed at different times, and the execution order of these sub-steps or stages is not necessarily sequential, but can be executed in rotation or alternation with other steps or at least part of the sub-steps or stages of other steps.

[0181] In one embodiment, as shown in Figure 5 An information recommendation device is provided, which comprises an acquisition module 502, a first calculation module 504, a matching module 506, a second calculation module 508, a determination module 510 and a recommendation module 512. Wherein:

[0182] The acquisition module 502 is configured to acquire a plurality of application objects pre-classified in an application program;

[0183] The first calculation module 504 is configured to calculate a first upper confidence interval value of each application object according to first historical operation information of a first user related to each application object;

[0184] The matching module 506 is configured to determine at least one second user similar to the first user according to portrait information of the first user;

[0185] The second calculation module 508 is configured to calculate a second upper confidence interval value of each application object corresponding to each second user according to second historical operation information of the second user related to each application object;

[0186] The determination module 510 is configured to determine a target upper confidence interval value of each application object according to the first upper confidence interval value of each application object and the second upper confidence interval value of each application object corresponding to each second user;

[0187] The recommendation module 512 is configured to sort the application objects according to the target upper confidence interval values of the application objects, and recommend the first user with the sorted application objects.

[0188] In one embodiment, the determination module 510 comprises:

[0189] The first determination unit is configured to take any one of the application objects as a first application object.

[0190] The calculation unit is configured to calculate an average upper confidence interval value of the first application object according to the second upper confidence interval value of the first application object corresponding to each second user.

[0191] The second determination unit is configured to determine a target upper confidence interval value of the first application object according to the first upper confidence interval value of the first application object and the average upper confidence interval value of the first application object.

[0192] In one embodiment, the second determination unit is further configured to:

[0193] Obtain a first weight corresponding to the first upper confidence interval value of the first application object.

[0194] Determine a second weight corresponding to the average upper confidence interval value of the first application object according to the first weight; the sum of the first weight and the second weight is one.

[0195] Multiply the first weight and the first upper confidence interval value of the first application object to obtain a first weighted value of the first application object.

[0196] Multiply the second weight and the average upper confidence interval value of the first application object to obtain a second weighted value of the first application object.

[0197] Calculate the sum of the first weighted value and the second weighted value to obtain the target upper confidence interval value of the first application object.

[0198] In one embodiment, the second determination unit is further configured to:

[0199] Obtain a plurality of weight functions stored locally.

[0200] Take a weight function satisfying a preset condition as a target weight function.

[0201] Determine the first weight of the first application object according to the target weight function.

[0202] In one embodiment, the second determination unit is further configured to:

[0203] determine an evaluation function value corresponding to each of the weight functions respectively based on a preset evaluation function;

[0204] determine the weight function with the largest evaluation function value as a target weight function.

[0205] In an embodiment, the application object includes commodities, and the first calculation module 504 is further configured to:

[0206] determine the number of orders of each of the commodities by the first user according to the first historical operation information;

[0207] obtain historical recommendation data of recommending each of the commodities to the first user stored locally;

[0208] determine the total number of times of recommending all the commodities to the first user and the number of times of recommending each of the commodities to the first user according to the historical recommendation data;

[0209] determine a first upper limit confidence interval value of each of the commodities according to the number of orders of each of the commodities by the first user, the total number of times of recommending all the commodities to the first user, and the number of times of recommending each of the commodities to the first user;

[0210] wherein the first upper limit confidence interval value of each of the commodities is positively correlated with the number of orders of each of the commodities by the first user and the total number of times of recommending all the commodities to the first user, and is negatively correlated with the number of times of recommending each of the commodities to the first user.

[0211] In an embodiment, the first calculation module 504 is further configured to:

[0212] when the number of times of recommending each of the commodities to the first user is 0, determine that the first upper limit confidence interval value of the commodity with the number of times of recommendation of 0 is 0.

[0213] The information recommendation apparatus provided in the embodiments of the present application firstly acquires a plurality of application objects pre-classified in an application program, calculates first upper limit confidence interval values of the application objects according to first historical operation information of a first user related to each application object, determines at least one second user similar to the first user according to portrait information of the first user, calculates second upper limit confidence interval values of each application object corresponding to each second user according to second historical operation information of the second user related to each application object, then determines target upper limit confidence interval values of the application objects according to the first upper limit confidence interval values of the application objects and the second upper limit confidence interval values of each application object corresponding to each second user, and further sorts the application objects according to the target upper limit confidence interval values of the application objects and recommends the application objects to the first user. Since the UCB values corresponding to each application object are determined according to the UCB values of the same application object by the first user and similar users, the preferences of the user and the preferences of similar users are considered when information is recommended, the accuracy of information recommendation is ensured, and the recommendation of unpopular information and newly listed information can also be realized.

[0214] The specific limitations of the information recommendation apparatus can be referred to the limitations of the information recommendation method in the foregoing, and will not be described herein. Each module in the information recommendation apparatus can be realized by software, hardware and combinations thereof in whole or in part. Each module can be embedded in or independent of the processor in the computer device in hardware form, or can be stored in the memory in the device in software form, so as to be called and executed by the processor to perform the operations corresponding to each module.

[0215] In one embodiment, a device, which can be a server, has an internal structure diagram as shown in Figure 6 The device includes a processor, a memory and a network interface connected through a system bus. The processor of the device is used to provide computing and control capabilities. The memory of the device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operating system and the computer program in the non-volatile storage medium to run. The database of the device is used to store user registration information, historical operation records, historical recommendation data and other data. The network interface of the device is used to communicate with external terminals through network connection. The computer program is executed by the processor to implement an information recommendation method.

[0216] Those skilled in the art can understand that Figure 6 The skilled in the art can understand that the structure shown in the above is only a block diagram of part of the structure related to the scheme of the present application, and does not constitute a limitation on the device to which the scheme of the present application is applied. The specific device can include more or fewer components than those shown in the figure, or combine certain components, or have a different component arrangement.

[0217] In one embodiment, the information recommendation apparatus provided in the present application can be implemented in the form of a computer program, which can run on the device as shown in Figure 6 The memory of the device can store various program modules constituting the information recommendation apparatus, such as the obtaining module, the first calculating module, the matching module, the second calculating module, the determining module and the recommending module as shown in Figure 5 The computer program constituted by the various program modules makes the processor execute the steps in the information recommendation method of each embodiment of the present application described in the specification.

[0218] For example, Figure 6 The device can execute the step of obtaining the plurality of application objects pre-classified in the application program through the obtaining module in the information recommendation apparatus as shown in Figure 5 The device can execute the step of calculating the first upper confidence interval value of each application object according to the first historical operation information related to the application object of the first user through the first calculating module. The device can execute the step of determining at least one second user similar to the first user according to the portrait information of the first user through the matching module. The device can execute the step of calculating the second upper confidence interval value of each application object corresponding to each second user according to the second historical operation information related to the application object of the second user through the second calculating module. The device can execute the step of determining the target upper confidence interval value of each application object according to the first upper confidence interval value of each application object and the second upper confidence interval value of each application object corresponding to each second user through the determining module. The device can execute the step of ranking each application object according to the target upper confidence interval value of each application object and recommending the first user through the recommending module.

[0219] In one embodiment, a computer readable storage medium is provided, which stores a computer program, and the computer program is executed by the processor to implement the steps in each method embodiment described above.

[0220] In one embodiment, a device is provided, which includes a memory and a processor, the memory stores a computer program, and the processor executes the computer program to implement the steps in each method embodiment described above.

[0221] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer readable storage medium, and when the computer program is executed, the processes of the above-mentioned embodiments of the methods can be included. Any reference to memory, database or other medium used in each embodiment provided by the present application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory or optical memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration but not limitation, RAM is available in various forms, such as static random access memory (SRAM) and dynamic random access memory (DRAM).

[0222] Each technical feature of the above embodiments can be combined arbitrarily. In order to make the description simple, all possible combinations of each technical feature in the above embodiments are not described, but as long as the combination of the technical features does not exist, it should be considered as the scope of the present application.

[0223] The above embodiments only express several implementation manners of the present application, and the description is more specific and detailed, but it should not be understood as a limitation on the scope of the patent. It should be pointed out that for ordinary skilled in the art, without departing from the concept of the present application, a number of modifications and improvements can be made, which are all within the scope of the present application. Therefore, the scope of the patent of the present application should be subject to the appended claims.

Claims

1. An information recommendation method characterized by comprising: The application comprises the following steps: Obtain a plurality of application objects pre-classified in an application program, and calculate a first upper limit confidence interval value of each application object according to first historical operation information of a first user related to each application object; Determine a plurality of second users similar to the first user according to portrait information of the first user; Calculate a second upper limit confidence interval value of each application object corresponding to each second user according to second historical operation information of the second user related to each application object; Take any application object in each application object as a first application object; Calculate an average upper limit confidence interval value of the first application object according to the second upper limit confidence interval value of the first application object corresponding to each second user; Determine a target upper limit confidence interval value of the first application object according to the first upper limit confidence interval value of the first application object and the average upper limit confidence interval value of the first application object; Sort each application object according to the target upper limit confidence interval value of each application object and recommend the first user with the sorted application objects.

2. The method of claim 1, wherein, The step of determining the target upper limit confidence interval value of the first application object according to the first upper limit confidence interval value of the first application object and the average upper limit confidence interval value of the first application object comprises the following steps: Obtain a first weight corresponding to the first upper limit confidence interval value of the first application object; Determine a second weight corresponding to the average upper limit confidence interval value of the first application object according to the first weight; the sum of the first weight and the second weight is one; Multiply the first weight and the first upper limit confidence interval value of the first application object to obtain a first weighted value of the first application object; Multiply the second weight and the average upper limit confidence interval value of the first application object to obtain a second weighted value of the first application object; Calculate the sum of the first weighted value and the second weighted value to obtain the target upper limit confidence interval value of the first application object.

3. The method of claim 2, wherein, The step of obtaining the first weight corresponding to the first upper limit confidence interval value of the first application object comprises the following steps: Obtain a plurality of weight functions stored locally; Determine a target weight function that meets a preset condition; Determine the first weight of the first application object according to the target weight function.

4. The method of claim 3, wherein, The step of determining the target weight function that meets the preset condition comprises the following steps: Determine an evaluation function value corresponding to each weight function based on a preset evaluation function; Determine the target weight function as the weight function with the maximum evaluation function value.

5. The method according to any one of claims 1 to 4, characterized in that, The application object comprises goods, and the step of calculating the first upper limit confidence interval value of each application object according to the first historical operation information of the first user related to each application object comprises the following steps: Determine the order placing frequency of the first user for each good according to the first historical operation information; Obtain historical recommendation data of the first user for each good stored locally; Determine the total frequency of recommending all goods to the first user and the recommendation frequency of each good to the first user according to the historical recommendation data; determine a first upper limit confidence interval value of each of the commodities according to the order number of each of the commodities by the first user, the total number of recommending all commodities to the first user, and the number of recommending each of the commodities to the first user; wherein the first upper limit confidence interval value of each of the commodities is positively correlated with the order number of each of the commodities by the first user, the total number of recommending all commodities to the first user, and negatively correlated with the number of recommending each of the commodities to the first user.

6. The method of claim 5, wherein, The method further comprises: when the number of recommending each of the commodities to the first user is 0, determining that the first upper limit confidence interval value of the commodity with the number of recommending being 0 is 0.

7. An information recommendation device characterized by comprising: comprise: an acquisition module configured to acquire a plurality of application objects pre-classified in an application program; a first calculation module configured to calculate a first upper limit confidence interval value of each of the application objects according to first historical operation information related to each of the application objects by a first user; a matching module configured to determine a plurality of second users similar to the first user according to portrait information of the first user; a second calculation module configured to calculate a second upper limit confidence interval value of each of the application objects corresponding to each of the second users according to second historical operation information related to each of the application objects by the second users; a determination module configured to determine a target upper limit confidence interval value of each of the application objects according to the first upper limit confidence interval value of each of the application objects and the second upper limit confidence interval value of each of the application objects corresponding to each of the second users; a recommendation module configured to sort each of the application objects according to the target upper limit confidence interval value of each of the application objects and recommend each of the application objects to the first user; the determination module comprises: a first determination unit configured to take any of the application objects as a first application object; a calculation unit configured to calculate an average upper limit confidence interval value of the first application object according to the second upper limit confidence interval value of the first application object corresponding to each of the second users; a second determination unit configured to determine a target upper limit confidence interval value of the first application object according to the first upper limit confidence interval value of the first application object and the average upper limit confidence interval value of the first application object.

8. A computer device comprising a memory and a processor, the memory storing a computer program, characterized in that, The processor executes the computer program to implement the steps of the information recommendation method in any one of claims 1 to 6.

9. A computer readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to implement the steps of the information recommendation method in any one of claims 1 to 6.

Images