Package recommendation method and device, electronic equipment and readable medium

Abstract

The invention relates to a package recommendation method and device, electronic equipment and a readable medium. The method comprises the following steps: acquiring user data of each user and package data of each package; wherein the user data comprises user historical behavior data and user demand data; calculating the user similarity between every two users by using the user historical behavior data and the user demand data, and calculating the package similarity between every two packages by using the package data; according to the user similarity, determining a first recommended package set of each user, and according to the package similarity, determining a second recommended package set of each user; and fusing the first recommended package set and the second recommended package set to obtain a target recommended package list of each user. According to the scheme provided by the invention, the data sparse problem and the package cold start problem can be solved, more accurate personalized recommendation is realized, and the handling efficiency of marketing acceptance business personnel is improved.

Description

Technical Field

[0001] The present application relates to the field of communication technology, and in particular to a package recommendation method, device, electronic device and readable medium. Background Art

[0002] In order to achieve business growth and reduce user churn rate, operators have launched a variety of packages. Users can choose different packages according to their needs to meet their daily communication and Internet needs.

[0003] In the related technologies, the selection of packages and sales products mainly relies on manual experience and lacks intelligent tool assistance, which leads to low efficiency in the marketing acceptance process. At the same time, the expression of packages is the expression of the operator's business system, which is too professional for users and cannot be recommended in real time and accurately for users' personalized needs. In addition, the related technologies use user historical data for training, but the effect is poor when facing data sparseness or package cold start problems. Due to concerns about insufficient historical data samples, a large amount of historical data is used for training, which easily recommends some packages that have been on the market for a long time, resulting in low recommendation efficiency. Summary of the invention

[0004] In order to solve or partially solve the problems existing in the related technologies, the present application provides a package recommendation method, device, electronic device and readable medium, which can solve the data sparsity problem and the package cold start problem, achieve more accurate personalized recommendations, and improve the processing efficiency of marketing acceptance personnel.

[0005] The first aspect of the present application provides a package recommendation method, the method comprising:

[0006] Acquire user data of each user and package data of each package; wherein the user data includes user historical behavior data and user demand data;

[0007] Using the user historical behavior data and the user demand data, calculate the user similarity between two of the users, and using the package data, calculate the package similarity between two of the packages;

[0008] Determining a first recommended package set for each of the users according to the user similarities, and determining a second recommended package set for each of the users according to the package similarities;

[0009] The first recommended package set and the second recommended package set are integrated to obtain a target recommended package list for each of the users.

[0010] In one implementation, the obtaining of user data of each user includes:

[0011] Configure multiple business scenarios; the multiple business scenarios include scenarios recommended for home users, scenarios recommended for business users, and scenarios recommended for gaming users;

[0012] Determining a target business scenario to which each user belongs from the multiple business scenarios;

[0013] Obtaining a recommendation factor of each of the users in the target service scenario, and obtaining usage information of an existing number of each of the users in an operator;

[0014] The recommendation factor is used as user demand data, and the usage information is used as user historical behavior data to obtain user data.

[0015] In one embodiment, the user historical behavior data includes user portrait information; the user historical behavior data and the user demand data are used to calculate the user similarity between two users, including:

[0016] Using historical user behavior data corresponding to user groups belonging to the same target business scenario, constructing an initial user behavior matrix for each of the user groups; and,

[0017] Using user portrait information corresponding to user groups belonging to the same target business scenario, constructing an initial user portrait matrix for each user group; and,

[0018] Using user demand data corresponding to user groups belonging to the same target business scenario, construct an initial user demand matrix for each user group;

[0019] Based on the initial user behavior matrix, the initial user portrait matrix and the initial user demand matrix of the same user group, the user similarity between any two users in the same user group is calculated.

[0020] In one embodiment, the package data includes applicable user groups; and the step of using the package data to calculate the package similarity between two packages includes:

[0021] Using the package data of the same user group, constructing a package feature matrix for each user group;

[0022] Based on the package feature matrix of the same user group, the package similarity between any two packages applicable to the same user group is calculated.

[0023] In one embodiment, the calculating the user similarity between two users in the same user group based on the initial user behavior matrix, the initial user portrait matrix and the initial user demand matrix of the same user group includes:

[0024] The initial user behavior matrix, the initial user portrait matrix and the initial user demand matrix of the same user group are respectively configured with corresponding first weight values; wherein each eigenvalue in the initial user demand matrix has a corresponding second weight value;

[0025] performing weighted processing on each eigenvalue in the initial user behavior matrix according to the corresponding first weight value to obtain a target user behavior matrix of the same user group; and

[0026] According to the corresponding first weight value, weighting is performed on each eigenvalue in the initial user portrait matrix to obtain a target user portrait matrix of the same user group; and

[0027] Performing weighted processing on each eigenvalue in the initial user demand matrix according to the corresponding first weight value and the corresponding second weight value to obtain a target user demand matrix of the same user group;

[0028] The target user behavior matrix, the target user portrait matrix and the target user demand matrix are combined to obtain a user information matrix of the same user group;

[0029] The user similarity between any two users in the same user group is calculated based on each eigenvalue in the user information matrix of the same user group through a cosine similarity model.

[0030] In one embodiment, the package includes an initial user sample set and an initial test sample set, each initial user sample in the initial user sample set is a favorite package of each user, each initial test sample in the initial test sample set is a candidate package to be recommended to each user, and the candidate package is a package currently on sale by an operator; the package feature matrix based on the same user group, calculating the package similarity between two packages applicable to the same user group, includes:

[0031] Standardizing each eigenvalue corresponding to each of the initial user samples in the package feature matrix of the same user group to obtain a standardized target user sample set, and standardizing each eigenvalue corresponding to each of the initial test samples in the package feature matrix of the same user group to obtain a labeled target test sample set;

[0032] The distance between each target user sample in the target user sample set and each target test sample in the target test sample set is calculated by the Euclidean distance formula to obtain the package similarity between each preferred package and each candidate package applicable to the same user group.

[0033] In one implementation, the fusing of the first recommended package set and the second recommended package set to obtain a target recommended package list for each user includes:

[0034] Selecting the first N recommended packages from the first recommended package set to form a first recommended package list, and selecting the first N recommended packages from the second recommended package set to form a second recommended package list; wherein N is a positive integer;

[0035] Merging and removing duplicates of the first recommended package list and the second recommended package list belonging to the same user to obtain a third recommended package list;

[0036] The plurality of recommended packages in the third recommended package list of each user are sorted according to the launch time to obtain a target recommended package list of each user.

[0037] A second aspect of the present application provides a meal plan recommendation device, the device comprising:

[0038] A data acquisition module, used to acquire user data of each user and package data of each package; wherein the user data includes user historical behavior data and user demand data;

[0039] A similarity calculation module, used to calculate the user similarity between two users by using the user historical behavior data and the user demand data, and to calculate the package similarity between two packages by using the package data;

[0040] A recommended package determination module, configured to determine a first recommended package set for each of the users according to the user similarity, and to determine a second recommended package set for each of the users according to the package similarity;

[0041] The recommended package fusion module is used to fuse the first recommended package set and the second recommended package set to obtain a target recommended package list for each user.

[0042] A third aspect of the present application provides an electronic device, including:

[0043] Processor; and

[0044] The memory stores executable codes thereon, and when the executable codes are executed by the processor, the processor is caused to execute the method as described above.

[0045] A fourth aspect of the present application provides a computer-readable storage medium having executable code stored thereon. When the executable code is executed by a processor of an electronic device, the processor is caused to execute the method as described above.

[0046] The technical solution provided by this application may have the following beneficial effects:

[0047] The solution provided by the present application obtains the user data of each user and the package data of each package; wherein the user data includes the user historical behavior data and the user demand data; the user historical behavior data and the user demand data are used to calculate the user similarity between two users, and the package data is used to calculate the package similarity between two packages; according to the user similarity, the first recommended package set of each user is determined, and according to the package similarity, the second recommended package set of each user is determined; the first recommended package set and the second recommended package set are integrated to obtain the target recommended package list of each user. The present application can solve the problem that the simple data description cannot allow users to quickly judge whether the package meets their needs by collecting user demand data in real time, and the recommendation based on the user's collaborative filtering algorithm can solve the problem of data sparsity, making the recommendation more accurate, and the recommendation based on the content recommendation algorithm can find potential target user groups for new packages in a short time, thereby solving the package cold start problem, and by combining user-based collaborative filtering and content-based recommendation, a hybrid recommendation strategy is formed, so that not only the similarity between users can be used for recommendation, but also the characteristics of the package itself can be used for recommendation, thereby achieving more accurate personalized recommendation and improving the handling efficiency of marketing acceptance business personnel.

[0048] It should be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] The above and other objects, features and advantages of the present application will become more apparent by describing in more detail the exemplary embodiments of the present application in conjunction with the accompanying drawings, wherein the same reference numerals generally represent the same components in the exemplary embodiments of the present application.

[0050] Figure 1 It is a flowchart of a method for recommending a meal package according to an embodiment of the present application;

[0051] Figure 2 is another flowchart of the meal plan recommendation method shown in an embodiment of the present application;

[0052] Figure 3 is a structural schematic diagram of a meal recommendation device shown in an embodiment of the present application;

[0053] Figure 4 It is a schematic diagram of the structure of an electronic device shown in an embodiment of the present application. DETAILED DESCRIPTION

[0054] The embodiments of the present application will be described in more detail below with reference to the accompanying drawings. Although the embodiments of the present application are shown in the accompanying drawings, it should be understood that the present application can be implemented in various forms and should not be limited by the embodiments described herein. On the contrary, these embodiments are provided to make the present application more thorough and complete, and to fully convey the scope of the present application to those skilled in the art.

[0055] The terms used in this application are for the purpose of describing specific embodiments only and are not intended to limit this application. The singular forms of "a", "said" and "the" used in this application and the appended claims are also intended to include plural forms unless the context clearly indicates other meanings. It should also be understood that the term "and / or" used herein refers to and includes any or all possible combinations of one or more associated listed items.

[0056] It should be understood that although the terms "first", "second", "third", etc. may be used in this application to describe various information, this information should not be limited to these terms. These terms are only used to distinguish the same type of information from each other. For example, without departing from the scope of this application, the first information may also be referred to as the second information, and similarly, the second information may also be referred to as the first information. Thus, the features defined as "first" and "second" may explicitly or implicitly include one or more of the features. In the description of this application, the meaning of "multiple" is two or more, unless otherwise clearly and specifically defined.

[0057] In the related art, the selection of packages and sales products mainly relies on manual experience and lacks intelligent tool assistance, resulting in low efficiency of the marketing acceptance process. At the same time, the expression of the package is the expression of the operator's business system, which is too professional for users and cannot be recommended in real time and accurately for the personalized needs of users. For example, in the problem of package retrieval, there is currently a lack of retrieval and analysis tools for package user dimensions. When handling business, business personnel can only search by sales product name or code, and recommend packages and sales products to users based on package descriptions (for example, the need to include more than 10G of traffic, more than 500 minutes of calls, etc.). For the real requirements of users (residence type: single, a family of two, a small family (3-4 people), a large family (5 people and above), the budget of members for package fees, whether members have specific usage habits, such as frequent use of social media, online videos or games, etc.), they mainly rely on experience to convert them into package attributes to search for sales products. This method relies on the experience of business personnel, has low business handling efficiency, and is difficult to support precision marketing for user needs.

[0058] In addition, related technologies use user historical data for training, but the effect is poor when facing data sparseness or package cold start problems. Due to concerns about insufficient historical data samples, a large amount of historical data is used for training, which easily leads to the recommendation of some packages that have been on the market for a long time, resulting in low recommendation efficiency.

[0059] In response to the above problems, the embodiments of the present application provide a package recommendation method. By collecting user demand data in real time, it can solve the problem that a simple data description cannot allow users to quickly determine whether a package meets their needs. The recommendation based on the user's collaborative filtering algorithm can solve the problem of data sparsity, making the recommendation more accurate. The recommendation based on the content recommendation algorithm can find potential target user groups for new packages in a short time, thereby solving the package cold start problem. By combining user-based collaborative filtering and content-based recommendation, a hybrid recommendation strategy is formed. In this way, not only can the similarity between users be used for recommendation, but also the characteristics of the package itself can be used for recommendation, thereby achieving more accurate personalized recommendations and improving the processing efficiency of marketing acceptance personnel.

[0060] The technical solution of the embodiments of the present application is described in detail below with reference to the accompanying drawings.

[0061] Figure 1 It is a flow chart of the package recommendation method shown in the embodiment of the present application.

[0062] See also Figure 1 , the package recommendation method of this application includes:

[0063] S110, obtaining user data of each user and package data of each package; wherein the user data includes user historical behavior data and user demand data.

[0064] In the embodiment of the present application, it can be applied to the recommendation system, and the recommendation system can obtain the user data of each user in the operator (such as a telecommunications operator), and the user data can include the user's historical behavior data and user demand data. Among them, the user's historical behavior data can be the use information of the user's existing number in the operator, for example, the user's historical behavior data can include at least one of the user's purchase record information, user portrait information and user consumption information in the operator, the purchase record information can refer to the user's past purchase behavior, for example, it can include the type of package purchased, the time of purchase, etc., the user portrait information can include the user's age, gender, occupation, geographical location and other basic information, and the user consumption information can include the user's call duration, traffic usage, SMS sending volume, etc. Among them, the user demand data can be the needs of the user's changes collected through surveys or real-time, for example, a list of demand parameters for the past year or the past two years can be collected, for example, the user demand data can include at least one of the family population, house area, and network speed requirements, the family population can refer to the number of family members of the user, the house area can refer to the user's living area, and the network speed requirements can refer to the user's demand for network speed.

[0065] The recommendation system can also obtain package data of each package currently on sale in an operator (such as a telecom operator). Operators will periodically launch new packages. Therefore, the package data obtained by the recommendation system may not only cover the package data of packages that users have purchased, but also cover the package data of new packages that operators periodically launch. Among them, the package data may include at least one of the package name, package price, package classification, applicable user group, package traffic, package duration, and package rate. The package name may refer to the specific name of the package, the package price may refer to the price of the package, the package classification may refer to the category to which the package belongs, the applicable user group may refer to which user group the package is suitable for (such as home users, business users, game users, etc.), the package traffic may refer to the Internet traffic included in the package, the package duration may refer to the call duration included in the package, and the package rate may refer to the Internet speed of the package.

[0066] S120, using the user historical behavior data and the user demand data to calculate the user similarity between any two users, and using the package data to calculate the package similarity between any two packages.

[0067] User data can be used as a data source for user-based collaborative filtering algorithms. User-based collaborative filtering algorithms (User CF) are a form of collaborative filtering algorithms that make recommendations based on similarities between users.

[0068] Package data can be used as a data source for content-based recommendation algorithms. Unlike collaborative filtering, content-based recommendation algorithms (CB) do not rely on the preferences of other users, but focus on analyzing the attributes of the items themselves.

[0069] Therefore, the recommendation system can use user historical behavior data and user demand data to calculate the user similarity between users, and can use package data to calculate the package similarity between packages.

[0070] S130, determining a first recommended package set for each user according to the user similarity, and determining a second recommended package set for each user according to the package similarity.

[0071] User-based collaborative filtering algorithms usually assume that "you are likely to like what people with similar preferences like". Therefore, the recommendation system can filter out users with similar preferences to each user based on user similarity, and then determine the first recommended package set for each user based on the packages purchased by users with similar preferences.

[0072] Content-based recommendation algorithms usually assume that "you are likely to like things similar to what you like", so the recommendation system can filter out similar packages of each user's favorite package based on package similarity, and then determine the second recommended package set for each user based on similar packages.

[0073] S140: Merge the first recommended package set and the second recommended package set to obtain a target recommended package list for each user.

[0074] The embodiment of the present application can combine user-based collaborative filtering and content-based recommendation results to generate a final recommendation list through a Top-N fusion strategy. In a specific implementation, after obtaining the first recommended package set for each user and the second recommended package set for each user, the recommendation system can merge the first recommended package set and the second recommended package set belonging to the same user together to provide the user with a more accurate and personalized target recommended package list.

[0075] The package recommendation method provided in the embodiment of the present application can achieve the following effects:

[0076] Improve user experience: Personalized package recommendations can better meet user needs and help users easily choose packages that suit them, thereby improving user experience and satisfaction.

[0077] Improve user stickiness: Providing personalized services to users can enhance their dependence and loyalty to operators (such as telecom operators), prompting users to use telecom services for a longer period of time.

[0078] Improve sales conversion rate: Through accurate personalized recommendations, users' recognition and willingness to purchase packages currently on sale by operators (such as telecom operators) can be increased, thereby improving sales conversion rate and promoting the sales of telecom packages.

[0079] As can be seen from the example, the scheme provided by the present application obtains the user data of each user and the package data of each package; wherein the user data includes the user historical behavior data and the user demand data; the user historical behavior data and the user demand data are used to calculate the user similarity between two users, and the package data is used to calculate the package similarity between two packages; according to the user similarity, the first recommended package set of each user is determined, and according to the package similarity, the second recommended package set of each user is determined; the first recommended package set and the second recommended package set are integrated to obtain the target recommended package list of each user. The present application can solve the problem that the simple data description cannot allow users to quickly judge whether the package meets their needs by collecting user demand data in real time, and the recommendation based on the user's collaborative filtering algorithm can solve the problem of data sparsity, making the recommendation more accurate, and the recommendation based on the content recommendation algorithm can find potential target user groups for new packages in a short time, thereby solving the package cold start problem, and by combining user-based collaborative filtering and content-based recommendation, a hybrid recommendation strategy is formed, so that not only the similarity between users can be used for recommendation, but also the characteristics of the package itself can be used for recommendation, thereby achieving more accurate personalized recommendation and improving the handling efficiency of marketing acceptance staff.

[0080] Figure 2 It is another flowchart of the meal plan recommendation method shown in this application.

[0081] See also Figure 2 , the package recommendation method of this application includes:

[0082] S210, obtaining user data of each user and package data of each package; wherein the user data includes user historical behavior data and user demand data.

[0083] The recommendation system can obtain user data of each user in an operator (e.g., a telecommunications operator), and the user data may include user historical behavior data and user demand data. Among them, user historical behavior data may be information on the use of a user's existing number in an operator, for example, user historical behavior data may include at least one of the user's purchase record information, user portrait information, and user consumption information in an operator, the purchase record information may refer to the user's past purchase behavior, for example, it may include the type of package purchased, the time of purchase, etc., the user portrait information may include basic information such as the user's age, gender, occupation, and geographic location, and the user consumption information may include the user's call duration, traffic usage, and SMS volume, etc. Among them, the user demand data may be the needs of users that are changed through surveys or real-time collection, for example, a list of demand parameters for the past year or the past two years may be collected, for example, the user demand data may include at least one of family population, house area, and network speed requirements, the family population may refer to the number of family members of the user, the house area may refer to the living area of ​​the user, and the network speed requirements may refer to the user's demand for network speed.

[0084] The recommendation system can also obtain package data of each package currently on sale in the operator (such as a telecom operator). The operator will periodically launch new packages. Therefore, the package data obtained by the recommendation system may not only cover the package data of the packages that the user has purchased, but also cover the package data of new packages that the operator launches periodically. Among them, the package data may include at least one of the package name, package price, package classification, applicable user group, package traffic, package duration and package rate. The package name may refer to the specific name of the package, the package price may refer to the price of the package, the package classification may refer to the category to which the package belongs, the applicable user group may refer to which user group the package is suitable for (such as family attributes, business attributes, game attributes, etc.), the package traffic may refer to the Internet traffic included in the package, the package duration may refer to the call duration included in the package, and the package rate may refer to the Internet speed of the package.

[0085] In one implementation, obtaining user data of each user may include:

[0086] Configure multiple business scenarios; the multiple business scenarios include home user recommendation scenarios, business user recommendation scenarios and game user recommendation scenarios; determine the target business scenario to which each user belongs from the multiple business scenarios; obtain the recommendation factor of each user in the target business scenario to which they belong, and obtain the usage information of each user's existing number in the operator; use the recommendation factor as user demand data, and use the usage information as user historical behavior data to obtain user data.

[0087] The embodiment of the present application can construct a collection method based on user personalized needs to obtain user demand data. In a specific implementation, in the initial stage, the embodiment of the present application can configure multi-dimensional recommendation factors, for example, multiple business scenarios can be configured, and the multiple business scenarios can include home user recommendation scenarios, business user recommendation scenarios, and game user recommendation scenarios. Then, different recommendation factors can be configured for different business scenarios, and corresponding values ​​can be set for each recommendation factor, so as to quickly and effectively collect the personalized needs of users.

[0088] For example, the recommended scenario for home users is: How many people are there in your family? How many phone numbers do you need? Do you use Internet TV at home? Do you often watch HD videos?

[0089] For example, business customers recommend scenarios such as: What is the nature of your work? Do you travel frequently? Do you need international roaming services?

[0090] For example, the recommended scenario for game users: What types of games do you mainly play? Do you have any special requirements for network speed?

[0091] In addition, the embodiment of the present application can also set a corresponding second weight value wn for each recommendation factor. For example, for a family user, family population and Internet TV usage may have a higher second weight value wn; for a business user, international roaming services and call duration may have a higher second weight value wn.

[0092] In the application stage, the recommendation system can determine the target business scenario to which each user belongs from the above-mentioned multiple business scenarios, and then obtain the recommendation factor of each user in the target business scenario to which they belong, so as to use the recommendation factor of each user as their own user demand data. For example, assuming that user A belongs to the user of the family user recommendation scenario (i.e., family user), and user B belongs to the user of the business customer recommendation scenario (i.e., business user), then their user demand data (i.e., recommendation factor) are different. For example, the user demand data of family user A may include family population, house area, and network speed requirements, etc., and the user demand data of business user B may include international roaming services and call duration, etc.

[0093] User historical behavior data may be the usage information of the user's existing number in the operator, so the recommendation system may directly obtain the usage information corresponding to each user's existing number from the operator (such as a telecom operator) so as to use the usage information corresponding to each user's existing number as their own user historical behavior data.

[0094] After obtaining the user demand data and the user historical behavior data of each user, the recommendation system can output the user demand data and the user historical behavior data belonging to the same user as the user data of the user.

[0095] S220, using the user historical behavior data and the user demand data to calculate the user similarity between any two users, and using the package data to calculate the package similarity between any two packages.

[0096] User data can be used as a data source for user-based collaborative filtering algorithms, and package data can be used as a data source for content-based recommendation algorithms. Therefore, the recommendation system can use user historical behavior data and user demand data to calculate the user similarity between users, and can use package data to calculate the package similarity between packages.

[0097] In one embodiment, the user historical behavior data includes user portrait information; using the user historical behavior data and the user demand data to calculate the user similarity between two users may include:

[0098] The initial user behavior matrix of each user group is constructed by using the user historical behavior data corresponding to the user groups belonging to the same target business scenario; and the initial user portrait matrix of each user group is constructed by using the user portrait information corresponding to the user groups belonging to the same target business scenario; and the initial user demand matrix of each user group is constructed by using the user demand data corresponding to the user groups belonging to the same target business scenario; based on the initial user behavior matrix, initial user portrait matrix and initial user demand matrix of the same user group, the user similarity between any two users in the same user group is calculated.

[0099] Different business scenarios correspond to different user demand data. Therefore, when calculating user similarity, the embodiment of the present application can be performed between two users belonging to the same target business scenario.

[0100] Therefore, the recommendation system can first classify each user into user groups. In a specific implementation, if the target business scenario of a user is a family user recommendation scenario, the recommendation system can classify the user into a user group with family attributes; if the target business scenario of a user is a business user recommendation scenario, the recommendation system can classify the user into a user group with business attributes; if the target business scenario of a user is a game user recommendation scenario, the recommendation system can classify the user into a user group with game attributes.

[0101] To facilitate the use of subsequent recommendation algorithms, the recommendation system can transform and preprocess the obtained data to construct mapping tables and different types of matrices.

[0102] Specifically, the recommendation system can assign a unique User ID to each user to achieve user-UserID mapping, and can assign a unique Offer ID to each package to achieve package-Offer ID mapping, and can assign a Feature to each feature value in the user data, and can assign a Feature to each feature value in the package data, thereby achieving feature value-Feature mapping. It should be noted that for non-numeric features (such as package names), one-hot encoding can be used to convert them into numerical form.

[0103] Afterwards, the recommendation system can use the user historical behavior data corresponding to the user groups belonging to the same target business scenario to construct the initial user behavior matrix of each user group. Specifically, the recommendation system can use the user historical behavior data corresponding to the user group with family attributes to construct the initial user behavior matrix of the user group with family attributes, and can use the user historical behavior data corresponding to the user group with business attributes to construct the initial user behavior matrix of the user group with business attributes, and can use the user historical behavior data corresponding to the user group with game attributes to construct the initial user behavior matrix of the user group with game attributes, as shown in Table 1:

[0104] UserID Feature1 Feature2 ... Features 1 F1b_1 F1b_2 ... F1b_n 2 F2b_1 F2b_2 ... F2b_n ... ... ... ... ... m Fmb_1 Fmb_2 ... Fmb_n

[0105] Table 1

[0106] Table 1 is the initial user behavior matrix of any user group. Taking the initial user behavior matrix of the user group with family attributes as an example, User ID 1 to User ID m represent family user 1 to family user m, respectively, and m is an integer greater than 2. Feature 1 to Feature n represent each feature value in the user historical behavior data corresponding to the user group with family attributes, respectively, and n is an integer greater than 2. For example, in Table 1, Feature 1 represents the purchase record information in the user historical behavior data, Feature 2 represents the user portrait information in the user historical behavior data, and Feature 3 represents the user consumption information in the user historical behavior data. Among them, the specific values ​​of Feature 1 to Feature n (F1b_1 to Fmb_n) are obtained by normalization calculation.

[0107] In order to construct a detailed user portrait, the recommendation system can extract user portrait information from the user's historical behavior data, so that the user portrait information corresponding to the user group belonging to the same target business scenario can be used to construct the initial user portrait matrix of each user group. Specifically, the recommendation system can use the user portrait information corresponding to the user group with family attributes to construct the initial user portrait matrix of the user group with family attributes, and can use the user portrait information corresponding to the user group with business attributes to construct the initial user portrait matrix of the user group with business attributes, and can use the user portrait information corresponding to the user group with game attributes to construct the initial user portrait matrix of the user group with game attributes, as shown in Table 2:

[0108] UserID Feature1 Feature2 ... Features 1 F1d_1 F1d_2 ... F1d_n 2 F2d_1 F2d_2 ... F2d_n ... ... ... ... ... m Fmd_1 Fmd_2 ... Fmd_n

[0109] Table 2

[0110] Table 2 is the initial user portrait matrix of any user group. Taking the initial user portrait matrix of the user group with family attributes as an example, User ID 1 to User ID m represent family user 1 to family user m, respectively, and m is an integer greater than 2. Feature 1 to Feature n represent each feature value in the user portrait information corresponding to the user group with family attributes, and n is an integer greater than 2. For example, in Table 2, Feature 1 represents the age in the user portrait information, Feature 2 represents the gender in the user portrait information, and Feature 3 represents the occupation in the user portrait information. Among them, the specific values ​​of Feature 1 to Feature n (F1d_1 to Fmd_n) are obtained by normalization calculation. For example, assuming that the maximum age in the user group with family attributes is 60 years old and the minimum age is 18 years old, the normalized value of a 35-year-old family user is: (35-18) / (60-18)≈0.357, which is approximately 0.3.

[0111] Similarly, the recommendation system can use the user demand data corresponding to the user groups belonging to the same target business scenario to construct the initial user demand matrix of each user group. Specifically, the recommendation system can use the user demand data corresponding to the user group with family attributes to construct the initial user demand matrix of the user group with family attributes, and can use the user demand data corresponding to the user group with business attributes to construct the initial user demand matrix of the user group with business attributes, and can use the user demand data corresponding to the user group with game attributes to construct the initial user demand matrix of the user group with game attributes, as shown in Table 3:

[0112]

[0113]

[0114] Table 3

[0115] Table 3 is the initial user demand matrix of any user group. Taking the initial user portrait matrix of the user group with family attributes as an example, User ID 1 to User ID m represent family user 1 to family user m, respectively, and m is an integer greater than 2. Feature 1 to Feature n represent each feature value in the user demand data corresponding to the user group with family attributes, and n is an integer greater than 2. For example, in Table 3, Feature 1 represents the family population in the user demand data, Feature 2 represents the house area in the user demand data, and Feature 3 represents the network speed requirement in the user demand data. Among them, the specific values ​​of Feature 1 to Feature n (F1n_1 to Fmn_n) are obtained by normalization calculation.

[0116] After constructing the initial user behavior matrix, the initial user portrait matrix, and the initial user demand matrix, respectively, the recommendation system can calculate the user similarity between two users in the same user group based on the initial user behavior matrix, the initial user portrait matrix, and the initial user demand matrix of the same user group. Exemplarily, assuming that users A1 to A3 are family users and users B1 to B3 are business users, the recommendation system can calculate the user similarity between family users A1 and A2, the user similarity between family users A1 and A3, and the user similarity between family users A2 and A3 based on the initial user behavior matrix, the initial user portrait matrix, and the initial user demand matrix of the user group with family attributes. In addition, the recommendation system can calculate the user similarity between business users B1 and B2, the user similarity between business users B1 and B3, and the user similarity between business users B2 and B3 based on the initial user behavior matrix, the initial user portrait matrix, and the initial user demand matrix of the user group with business attributes.

[0117] In one embodiment, based on the initial user behavior matrix, the initial user portrait matrix, and the initial user demand matrix of the same user group, calculating the user similarity between any two users in the same user group may include:

[0118] The initial user behavior matrix, the initial user portrait matrix and the initial user demand matrix of the same user group are respectively configured with corresponding first weight values; wherein each eigenvalue in the initial user demand matrix has a corresponding second weight value; according to the corresponding first weight value, each eigenvalue in the initial user behavior matrix is ​​weighted to obtain a target user behavior matrix of the same user group; and, according to the corresponding first weight value, each eigenvalue in the initial user portrait matrix is ​​weighted to obtain a target user portrait matrix of the same user group; and, according to the corresponding first weight value and the corresponding second weight value, each eigenvalue in the initial user demand matrix is ​​weighted to obtain a target user demand matrix of the same user group; the target user behavior matrix, the target user portrait matrix and the target user demand matrix are merged to obtain a user information matrix of the same user group; and the user similarity between two users in the same user group is calculated based on each eigenvalue in the user information matrix of the same user group through a cosine similarity model.

[0119] The embodiment of the present application can merge the initial user behavior matrix, the initial user portrait matrix and the initial user demand matrix into a user information matrix according to a certain weight ratio. In a specific implementation, the recommendation system can respectively configure corresponding first weight values ​​for the initial user behavior matrix, the initial user portrait matrix and the initial user demand matrix of the same user group. Exemplarily, for a user group with family attributes, the first weight value Wb corresponding to the initial user behavior matrix can be configured as 0.3, the first weight value Wd corresponding to the initial user portrait matrix can be configured as 0.3, and the first weight value Wn corresponding to the initial user demand matrix can be configured as 0.4.

[0120] Since a corresponding second weight value wn has been set for each recommendation factor in the initial stage, each eigenvalue in the initial user demand matrix has a corresponding second weight value wn.

[0121] Afterwards, the recommendation system can perform weighted processing on each eigenvalue in the initial user behavior matrix according to the corresponding first weight value Wb, and can obtain the target user behavior matrix of the same user group. Specifically, the recommendation system can perform weighted processing on each eigenvalue in the initial user behavior matrix according to the first weight value Wb corresponding to the initial user behavior matrix of the user group with family attributes, and can obtain the target user behavior matrix of the user group with family attributes, and can perform weighted processing on each eigenvalue in the initial user behavior matrix according to the first weight value Wb corresponding to the initial user behavior matrix of the user group with business attributes, and can obtain the target user behavior matrix of the user group with business attributes, and can perform weighted processing on each eigenvalue in the initial user behavior matrix according to the first weight value Wb corresponding to the initial user behavior matrix of the user group with game attributes, and can obtain the target user behavior matrix of the user group with game attributes.

[0122] Similarly, the recommendation system can perform weighted processing on each eigenvalue in the initial user portrait matrix according to the corresponding first weight value Wd, and obtain the target user portrait matrix of the same user group. Specifically, the recommendation system can perform weighted processing on each eigenvalue in the initial user portrait matrix according to the first weight value Wd corresponding to the initial user portrait matrix of the user group with family attributes, and obtain the target user portrait matrix of the user group with family attributes, and can perform weighted processing on each eigenvalue in the initial user portrait matrix according to the first weight value Wd corresponding to the initial user portrait matrix of the user group with business attributes, and obtain the target user portrait matrix of the user group with business attributes, and can perform weighted processing on each eigenvalue in the initial user portrait matrix according to the first weight value Wd corresponding to the initial user portrait matrix of the user group with game attributes, and obtain the target user portrait matrix of the user group with game attributes.

[0123] Similarly, the recommendation system can perform weighted processing on each eigenvalue in the initial user demand matrix according to the corresponding first weight value Wn and the corresponding second weight value wn, and obtain the target user demand matrix of the same user group. Specifically, the recommendation system can perform weighted processing on each eigenvalue in the initial user demand matrix according to the first weight value Wn and the corresponding second weight value wn corresponding to the initial user demand matrix of the user group with family attributes, and obtain the target user demand matrix of the user group with family attributes, and can perform weighted processing on each eigenvalue in the initial user demand matrix according to the first weight value Wn and the corresponding second weight value wn corresponding to the initial user demand matrix of the user group with business attributes, and obtain the target user demand matrix of the user group with business attributes, and can perform weighted processing on each eigenvalue in the initial user demand matrix according to the first weight value Wn and the corresponding second weight value wn corresponding to the initial user demand matrix of the user group with game attributes, and obtain the target user demand matrix of the user group with game attributes.

[0124] It should be noted that the first weight values ​​of the same type of matrices in different business scenarios can be the same. For example, in different business scenarios, the first weight values ​​corresponding to the initial user behavior matrix, the initial user portrait matrix, and the initial user demand matrix can all be set in the ratio of [0.3, 0.3, 0.4].

[0125] After obtaining the target user behavior matrix, target user portrait matrix and target user demand matrix respectively, the recommendation system can merge the target user behavior matrix, target user portrait matrix and target user demand matrix to obtain the user information matrix of the same user group, as shown in Table 4:

[0126]

[0127] Table 4

[0128] Table 4 is the user information matrix of any user group. Taking the user information matrix of the user group with family attributes as an example, Demographic is the target user portrait matrix of the user group with family attributes, Behavioral is the target user behavior matrix of the user group with family attributes, and Needs is the target user need matrix of the user group with family attributes.

[0129] Afterwards, the recommendation system can call the cosine similarity model to calculate the user similarity between two users in the same user group based on each eigenvalue in the user information matrix of the same user group using the cosine similarity model. The specific calculation formula is as follows:

[0130]

[0131] Wherein, (A) and (B) respectively represent the vector representation of two users in the same user group on all the features in Table 4, and (n) is the total number of features in Table 4, for example, (n)=3*n.

[0132] According to the above calculation formula, the user similarity between two users in the same user group is calculated respectively, and then all the calculation results of the same user group can be used to construct the user similarity matrix of the same user group, as shown in Table 5:

[0133] UserID 1 2 ... m 1 1.0 0.8 ... 0.6 2 0.8 1.0 ... 0.7 ... ... ... ... ... m 0.6 0.7 ... 1.0

[0134] Table 5

[0135] Table 5 is the user similarity matrix of any user group. Taking the user similarity matrix of the user group with family attributes as an example, for family user 1 (i.e., the family user with User ID 1), "1" represents his own user similarity. He and himself are completely similar, so the value of "1" is 1.0, "2" represents the user similarity between him and family user 2, ..., m represents the user similarity between him and family user m.

[0136] In one embodiment, the package data includes applicable user groups; using the package data, calculating the package similarity between two packages may include:

[0137] The package data of the same user group is used to construct the package feature matrix of each user group; based on the package feature matrix of the same user group, the package similarity between any two packages applicable to the same user group is calculated.

[0138] Different user groups have different feature values ​​of package data, so the recommendation system can extract the applicable user groups from the package data, and then classify all packages according to the user groups that each package is applicable to. In the specific implementation, if a package is suitable for a user group with family attributes, then the package can be classified as a family package type; if a package is suitable for a user group with business attributes, then the package can be classified as a business package type; if a package is suitable for a user group with gaming attributes, then the package can be classified as a gaming package type.

[0139] Similarly, in order to facilitate the use of subsequent recommendation algorithms, the recommendation system can convert and preprocess the obtained data to construct a package feature matrix.

[0140] Afterwards, the recommendation system can use the package data of the same user group to construct a package feature matrix for each user group. Specifically, the recommendation system can use the package data of the family package type to construct a package feature matrix for the user group with family attributes, and can use the package data of the business package type to construct a package feature matrix for the user group with business attributes, and can use the package data of the game package type to construct a package feature matrix for the user group with game attributes, as shown in Table 6:

[0141]

[0142]

[0143] Table 6

[0144] Table 6 is a package feature matrix of any user group. Taking the package feature matrix of the user group with family attributes as an example, Offer ID 1 to Offer ID N represent family package 1 to family package N, respectively, and N is an integer greater than 2. Feature1 to Feature n represent each feature value in the package data of the family package type, and n is an integer greater than 2. For example, in Table 6, Feature 1 represents the package name in the package data, Feature 2 represents the package price in the package data, and Feature 3 represents the package category in the package data. Among them, the specific values ​​of Feature 1 to Feature n (F1t_1 to FNt_n) are obtained by normalization calculation.

[0145] After constructing the package feature matrix, the recommendation system can calculate the package similarity between the packages applicable to the same user group based on the package feature matrix of the same user group. For example, assuming that packages A1 to A3 belong to the family package type, and packages B1 to B3 belong to the business package type, the recommendation system can calculate the package similarity between family package A1 and family package A2, the package similarity between family package A1 and family package A3, and the package similarity between family package A2 and family package A3 based on the package feature matrix of the user group with family attributes. In addition, the recommendation system can calculate the package similarity between business package B1 and business package B2, the package similarity between business package B1 and business package B3, and the package similarity between business package B2 and business package B3 based on the package feature matrix of the user group with business attributes.

[0146] In one embodiment, the package includes an initial user sample set and an initial test sample set, each initial user sample in the initial user sample set is a favorite package of each user, each initial test sample in the initial test sample set is a candidate package to be recommended to each user, and the candidate package is a package currently on sale by the operator; based on the package feature matrix of the same user group, calculating the package similarity between two packages applicable to the same user group may include:

[0147] The eigenvalues ​​corresponding to the initial user samples in the package feature matrix of the same user group are standardized to obtain a standardized target user sample set, and the eigenvalues ​​corresponding to the initial test samples in the package feature matrix of the same user group are standardized to obtain a labeled target test sample set; the distance between each target user sample in the target user sample set and each target test sample in the target test sample set is calculated by the Euclidean distance formula to obtain the package similarity between each preferred package and each candidate package applicable to the same user group.

[0148] The recommendation system can determine each user's favorite package from the packages that each user has purchased, so as to output each user's favorite package as each initial user sample in the initial user sample set X. In addition, the recommendation system can sort multiple packages currently on sale by operators (such as telecom operators) according to their launch time, and each package can be used as a candidate package to be recommended to each user, so as to output each user's candidate package as each initial test sample in the initial test sample set T.

[0149] In order to ensure that the scales of different eigenvalues ​​are consistent, the embodiments of the present application can use the Z-Score (standard score) method to standardize each eigenvalue corresponding to each initial user sample in the package feature matrix of the same user group, and obtain a standardized target user sample set, and standardize each eigenvalue corresponding to each initial test sample in the package feature matrix of the same user group, and obtain a labeled target test sample set.

[0150] Among them, the calculation formula of the Z-Score (standard score) method is as follows:

[0151]

[0152] Among them, X' is the standardized eigenvalue, X is the original eigenvalue, μ is the mean, and σ is the standard deviation.

[0153] In one example, assuming that users A to C are family users, an initial user sample set X of family users A to C can be obtained:

[0154] The eigenvalues ​​corresponding to the initial user sample (favorite package) X_A of household user A are:

[0155] X_A = [Internet speed: 1000M, traffic: 100G, duration: 1000 minutes, text messages: 200, consumption: 300 yuan].

[0156] The eigenvalues ​​corresponding to the initial user sample (favorite package) X_B of household user B are:

[0157] X_B = [Internet speed: 500M, traffic: 50G, duration: 800 minutes, text messages: 100, consumption: 200 yuan].

[0158] The eigenvalues ​​corresponding to the initial user sample (favorite package) X_C of household user C are:

[0159] X_C = [Internet speed: 300M, traffic: 40G, duration: 300 minutes, text messages: 10, consumption: 100 yuan].

[0160] In addition, assuming that Package 1 to Package 3 belong to the family package type, the initial test sample set T of Family Package 1 to Family Package 3 can be obtained:

[0161] The eigenvalues ​​corresponding to the initial test sample (candidate package) T_1 of family package 1 are:

[0162] T_1 = [Internet speed: 1000M, traffic: 100G, duration: 1200 minutes, text messages: 200, monthly rent: 299 yuan].

[0163] The eigenvalues ​​corresponding to the initial test sample (candidate package) T_2 of family package 2 are:

[0164] T_2 = [Internet speed: 500M, traffic: 60G, duration: 600 minutes, text messages: 100, monthly rent: 199 yuan].

[0165] The eigenvalues ​​corresponding to the initial test sample (candidate package) T_3 of family package 3 are:

[0166] T_3 = [Internet speed: 300M, traffic: 40G, duration: 300 minutes, text messages: 50, monthly rent: 99 yuan].

[0167] The specific values ​​of each characteristic value in the above examples are obtained from the package characteristic matrix of the user group with family attributes.

[0168] For the initial user samples (preferred packages) of family users A to C and the initial test sample (candidate package) of family package 1, the mean μ and the corresponding standard deviation σ of network speed (Speed), traffic (Data), duration (Duration), text messages (SMS), and consumption (Cost) are calculated respectively.

[0169] Taking the network speed (Speed) as an example, calculate the mean value μ corresponding to the network speed (Speed) speed and the corresponding standard deviation σ speed :

[0170]

[0171] According to the above calculation method, the means and standard deviations of data, duration, SMS, and cost are obtained respectively:

[0172] μ Data =60,μ Duration =700, μ SMS =100, μ Cost =199.17;

[0173] σ Data =20,σ Duration =326.59,σ SMS =57.74,σ Cost =72.04;

[0174] Taking the initial user sample (favorite package) X_A as an example, the initial user sample (favorite package) X_A is standardized to obtain its standardized eigenvalue X' A :

[0175]

[0176] According to the above calculation method, the standardized target user sample set and the annotated target test sample set are obtained respectively:

[0177] Standardized target user sample set:

[0178] Target user sample of household user A (preferred package): X_A' = [1.26, 1, 0.92, 1.73, 1.39];

[0179] Target user sample of household user A (preferred package): X_B' = [-0.46, 0, -0.31, 0, 0.01];

[0180] Target user sample of household user C (preferred package): X_C' = [-1.15, -1, -1.22, -0.87, -1.38].

[0181] Standardized target test sample set:

[0182] Target test sample of Family Package 1 (candidate package): T_1' = [1.26, 2, 1.53, 1.73, 1.38];

[0183] Target test sample of Family Package 2 (candidate package): T_2' = [-0.46, 1, -0.31, 0, 0.01];

[0184] Target test sample of family package 3 (candidate package): T_3' = [-1.15, -1, -1.22, -0.87, -1.38].

[0185] The embodiment of the present application can use the Euclidean distance formula to measure the similarity between the target user sample and the target test sample:

[0186]

[0187] Among them, a r (x i ) represents the rth eigenvalue of a target user sample i, a r (x j ) represents the rth eigenvalue of a target test sample j.

[0188] Taking the target user sample (favorite package) X_A' as an example, calculate the distance between the target user sample (favorite package) X_A' and the target test samples (candidate packages) T_1', T_2' and T_3' in the target test sample set:

[0189]

[0190] d(X' A ,T′2)≈3.06;

[0191] d(X' A ,T′3)≈5.37.

[0192] According to the above calculation method, the distances between the target user sample (preferred package) X_B' and the target test samples (candidate packages) T_1', T_2' and T_3' in the target test sample set can be obtained, and the distances between the target user sample (preferred package) X_C' and the target test samples (candidate packages) T_1', T_2' and T_3' in the target test sample set can be obtained, thereby obtaining the package similarity between each preferred package and each candidate package of the user group suitable for family attributes.

[0193] According to the above calculation method, the package similarity between each preferred package of the user group with business attributes and each candidate package can be obtained, and the package similarity between each preferred package of the user group with gaming attributes and each candidate package can be obtained.

[0194] S230, determining a first recommended package set for each user according to the user similarity, and determining a second recommended package set for each user according to the package similarity.

[0195] User-based collaborative filtering algorithms usually assume that "you are likely to like what people with similar preferences like", so in the user similarity matrix (Table 5) of the same user group, if the user similarity between a user and other users is greater than the similarity threshold (for example, 0.8), these users can be considered to be users with similar preferences to the user. Then, the first recommended package set for the user can be determined based on the packages purchased by users with similar preferences.

[0196] In one example, assuming that the user similarities between family user A and family user B1, family user B2, family user B3 and family user B4 are 0.9, 0.8, 0.7 and 0.8 respectively, then based on the similarity threshold (0.8), it can be screened out that users with similar preferences to family user A are family user B1, family user B2 and family user B4, and the recommendation system can obtain the packages purchased by these users with similar preferences. Assuming that the package purchased by family user B1 is [offerID 1, offerID 2], the package purchased by family user B2 is [offerID 2, offerID 5], and the package purchased by family user B4 is [offerID 5, offerID N], it can be determined that the first recommended package set for family user A is [offerID 1, offerID 2, offerID 5, offerIDN].

[0197] To facilitate the subsequent execution of Top-N fusion, after determining the first recommended package set, the recommendation system can calculate the recommendation score (User-Based CF Score) of each recommended package in the first recommended package set. The User-Based CF Score can be calculated based on the number of appearances of the recommended package in the packages purchased by users with similar preferences and the user similarity corresponding to the recommended package. The recommendation system can sort each recommended package in the first recommended package set based on the User-Based CF Score, and then execute the next step S240.

[0198] Following the above example, after determining the first recommended package set [offerID 1, offerID 2, offerID 5, offerIDN] for family user A, the recommendation scores (User-BasedCF Score) corresponding to offerID 1, offerID 2, offerID 5, offerIDN can be calculated respectively:

[0199] OfferID 1: User-BasedCF Score=0.9*1+0.8*0+0.8*0=0.9;

[0200] OfferID 2: User-BasedCF Score=0.9*1+0.8*1+0.8*0=1.7;

[0201] OfferID 5: User-BasedCF Score=0.9*0+0.8*1+0.8*1=1.6;

[0202] OfferIDN: User-BasedCF Score=0.9*0+0.8*0+0.8*1=0.8;

[0203] Each recommended package in the first recommended package set is sorted based on the User-BasedCF Score to obtain the first recommended package set recommended to the family user A based on user-based collaborative filtering: [offerID 2, offerID 5, offerID 1, offerIDN].

[0204] Content-based recommendation algorithms usually assume that "you are likely to like things that are similar to what you like". Therefore, after calculating the distance between each target user sample in the target user sample set and each target test sample in the target test sample set using the Euclidean distance formula, and obtaining the package similarity between each preferred package and each candidate package applicable to the same user group, the recommendation system can use the K-nearest neighbor algorithm to sort these package similarities (i.e., distances) from small to large, and then select the first k target test samples with the smallest distance as the neighbors of the current target user sample, and predict the recommendation status of the current target user sample based on the preferences of these neighbors. Among them, the k value should be a relatively small value, and the optimal k value can be selected through cross-validation.

[0205] In one example, according to the distances between the target test sample (favorite package) X_A' of the family user A and the target test samples (candidate packages) T_1', T_2' and T_3' in the target test sample set: d(X' A ,T′1)≈1.17, d(X' A ,T′2)≈3.06, d(X' A ,T′3)≈5.37, it can be determined that the distance between the target test sample (favorite package) X_A' of family user A and the target test sample (candidate package) T_1' is the smallest (1.17), that is, the package similarity between the two is the smallest, followed by the target test sample (candidate package) T_2'(3.06), and finally the target test sample (candidate package) T_3'(5.37). Therefore, it can be considered that the target test sample (candidate package) T_1' is closest to the preference of family user A, so the recommendation priority is the highest, so it can be determined that the second recommended package set for family user A is [offerID 1,offerID 2,offerID 3].

[0206] S240, selecting the first N recommended packages from the first recommended package set to form a first recommended package list, and selecting the first N recommended packages from the second recommended package set to form a second recommended package list; wherein N is a positive integer.

[0207] The embodiment of the present application can combine user-based collaborative filtering and content-based recommendation results to generate a final recommendation list through a Top-N fusion strategy. In a specific implementation, after obtaining the first recommended package set of each user and the second recommended package set of each user, the recommendation system can select the first N recommended packages from the first recommended package set of the current user, and the first N recommended packages can form the first recommended package list of the user, and select the first N recommended packages from the second recommended package set of the current user, and the first N recommended packages can form the second recommended package list of the user, where N is a positive integer, for example, the value range of N can be [1,10] or [1,5], which is not limited in the embodiment of the present application.

[0208] In one example, assuming that the first recommended package set of family user A is [offerID 2, offerID 5, offerID 1, offerID N], and the second recommended package set of family user A is [offerID 1, offerID 2, offerID 3], and N is 2, the recommendation system can select the first two recommended packages from the first recommended package set to form the first recommended package list of family user A [offerID 2, offerID 5], and can select the first two recommended packages from the second recommended package set to form the second recommended package list of family user A [offerID 1, offerID 2].

[0209] S250: Merge and remove duplicates of the first recommended package list and the second recommended package list belonging to the same user to obtain a third recommended package list.

[0210] The recommendation system may merge the first recommended package list and the second recommended package list belonging to the same user. Since there may be duplicate recommended packages in the first recommended package list and the second recommended package list, they may be deduplicated to remove the duplicate recommended packages, thereby obtaining a third recommended package list.

[0211] In one example, assuming that the first recommended package list of family user A is [offerID 2, offerID 5], and the second recommended package list of family user A is [offerID 1, offerID 2], the recommendation system can merge the two lists. Since there are repeated recommended packages offerID 2 in the two lists, one of the offerID2s is removed, so that the third recommended package list can be obtained: [offerID 1, offerID 2, offerID 5].

[0212] S260, sorting the multiple recommended packages in the third recommended package list of each user according to the launch time to obtain a target recommended package list of each user.

[0213] Operators will periodically launch new packages. In order to reduce the cold start time of new packages, the recommendation system can sort multiple recommended packages in each user's third recommended package list according to the launch time, so as to obtain a target recommended package list for each user.

[0214] In one example, assume that the third recommended package list for home user A is [offerID 1, offerID 2, offerID 5], where offerID1 and offerID2 are new packages launched by the operator in recent times, that is, offerID2 is launched the latest, followed by offerID 1, and then offerID 5. Therefore, the recommendation system sorts the multiple recommended packages in the third recommended package list of home user A according to the launch time, and obtains the target recommended package list for home user A: [offerID 2, offerID 1, offerID 5].

[0215] Furthermore, after obtaining the target recommended package list for each user, it can also be screened in combination with actual business needs to ensure that the final recommendation results meet specific conditions, wherein the specific conditions may include at least one of applicable area conditions, authority restriction conditions and target user group conditions, wherein the applicable area conditions may refer to checking whether each recommended package in the target recommended package list is applicable to the current user's geographic location, the authority restriction conditions may refer to determining whether the current user is eligible to order certain special packages, and the target user group conditions may refer to verifying whether each recommended package in the target recommended package list is designed for a specific group.

[0216] After the above steps of screening, a target recommended package list for each user is finally formed.

[0217] The package recommendation method provided in the embodiment of the present application can achieve the following effects:

[0218] (1) Economic Benefits

[0219] Increase sales: Through accurate personalized recommendations, users’ purchase rate can be increased. The recommendation system can recommend the most suitable packages based on the user’s historical behavior and preferences, thereby increasing the conversion rate.

[0220] Reduce customer churn: By providing packages that meet user needs, you can improve user satisfaction, reduce the number of customers switching to competitors due to dissatisfaction, and reduce customer churn rate.

[0221] (2) User satisfaction

[0222] Improve user experience: Personalized recommendations can significantly improve user experience and make users feel more attentive. It is easier for users to find packages that meet their needs, reducing the problem of choice difficulties.

[0223] Enhance user stickiness: By continuously providing high-quality recommendation services, user loyalty can be enhanced, making them more willing to use telecommunications products and services for a long time.

[0224] (3) Operational efficiency

[0225] Automated processing: The recommendation system can automatically train and output a large amount of data analysis and recommendation tasks, reducing manual pressure and improving operational efficiency.

[0226] It can be seen from this example that the solution provided by this application can solve the problem that simple data description cannot allow users to quickly judge whether the package meets their needs by collecting user demand data in real time, and the recommendation based on the user's collaborative filtering algorithm can solve the problem of data sparsity, making the recommendation more accurate, and the content-based recommendation algorithm can find potential target user groups for new packages in a short time, thereby solving the package cold start problem, and by combining user-based collaborative filtering and content-based recommendation, a hybrid recommendation strategy is formed, which can not only use the similarity between users for recommendation, but also use the characteristics of the package itself for recommendation, thereby achieving more accurate personalized recommendations and improving the processing efficiency of marketing acceptance personnel.

[0227] Furthermore, the solution provided in the present application, through humanized configuration of recommendation factors and collection of daily life needs, can solve the problem that simple data description cannot allow users to quickly judge whether the package meets their needs, help users easily choose a package that suits them, thereby improving user experience and satisfaction, and can enhance users' dependence and loyalty to telecommunications, prompting users to use telecommunications services for a longer period of time.

[0228] Furthermore, the solution provided in the present application is that for newly launched packages, the recommendation system can use a content-based recommendation method to match the package characteristics (such as traffic, duration, rate, etc.) with the user's preferred packages, thereby finding potential target user groups for the new package in a short time and reducing the cold start time of the new package.

[0229] Furthermore, the solution provided in the present application forms a hybrid recommendation strategy by combining user-based collaborative filtering and content-based recommendation. This approach can not only make recommendations based on the similarities between users, but also make recommendations based on the characteristics of the package itself, thereby achieving more accurate personalized recommendations, thereby increasing users' recognition and willingness to purchase telecommunications packages, improving sales conversion rates, and promoting the sales of telecommunications packages.

[0230] Corresponding to the aforementioned application function implementation method embodiment, the present application also provides a package recommendation device, an electronic device, a computer-readable storage medium and corresponding embodiments.

[0231] Figure 3 It is a structural schematic diagram of a meal plan recommendation device shown in an embodiment of the present application.

[0232] See also Figure 3 , the present application provides a package recommendation device, which may include:

[0233] The data acquisition module 310 is used to acquire user data of each user and package data of each package; wherein the user data includes user historical behavior data and user demand data;

[0234] A similarity calculation module 320 is used to calculate the user similarity between two users by using the user historical behavior data and the user demand data, and to calculate the package similarity between two packages by using the package data;

[0235] The recommended package determination module 330 is used to determine a first recommended package set for each user according to user similarity, and to determine a second recommended package set for each user according to package similarity;

[0236] The recommended package fusion module 340 is used to fuse the first recommended package set and the second recommended package set to obtain a target recommended package list for each user.

[0237] In one embodiment, the data acquisition module 310 may include:

[0238] The business scenario configuration submodule is used to configure various business scenarios; the various business scenarios include scenarios recommended for home users, scenarios recommended for business users, and scenarios recommended for gaming users;

[0239] A target business scenario determination submodule is used to determine the target business scenario to which each user belongs from a variety of business scenarios;

[0240] The data acquisition submodule is used to obtain the recommendation factor of each user in the target service scenario, and to obtain the usage information of each user's existing number in the operator;

[0241] The data definition submodule is used to obtain user data by taking the recommendation factor as user demand data and the usage information as user historical behavior data.

[0242] In one embodiment, the user historical behavior data includes user portrait information; the similarity calculation module 320 may include:

[0243] An initial user behavior matrix construction submodule is used to construct an initial user behavior matrix for each user group using user historical behavior data corresponding to the user group belonging to the same target business scenario; and

[0244] An initial user portrait matrix construction submodule is used to construct an initial user portrait matrix for each user group using user portrait information corresponding to the user group belonging to the same target business scenario; and

[0245] The initial user demand matrix construction submodule is used to construct the initial user demand matrix of each user group using the user demand data corresponding to the user groups belonging to the same target business scenario;

[0246] The user similarity calculation submodule is used to calculate the user similarity between any two users in the same user group based on the initial user behavior matrix, initial user portrait matrix and initial user demand matrix of the same user group.

[0247] In one embodiment, the package data includes applicable user groups; the similarity calculation module 320 may include:

[0248] The package feature matrix construction submodule is used to construct the package feature matrix of each user group using the package data of the same user group;

[0249] The package similarity calculation submodule is used to calculate the package similarity between any two packages applicable to the same user group based on the package feature matrix of the same user group.

[0250] In one embodiment, the user similarity calculation submodule may include:

[0251] A weight value configuration unit, used to configure corresponding first weight values ​​for the initial user behavior matrix, the initial user portrait matrix and the initial user demand matrix of the same user group, respectively; wherein each eigenvalue in the initial user demand matrix has a corresponding second weight value;

[0252] A first weighted processing unit is used to perform weighted processing on each eigenvalue in the initial user behavior matrix according to the corresponding first weight value to obtain a target user behavior matrix of the same user group; and

[0253] A second weighted processing unit is used to perform weighted processing on each eigenvalue in the initial user portrait matrix according to the corresponding first weight value to obtain a target user portrait matrix of the same user group; and

[0254] A third weighted processing unit, configured to perform weighted processing on each eigenvalue in the initial user demand matrix according to the corresponding first weight value and the corresponding second weight value, so as to obtain a target user demand matrix of the same user group;

[0255] A matrix merging unit, used to merge the target user behavior matrix, the target user portrait matrix and the target user demand matrix to obtain a user information matrix of the same user group;

[0256] The user similarity calculation unit is used to calculate the user similarity between two users in the same user group based on each eigenvalue in the user information matrix of the same user group through a cosine similarity model.

[0257] In one embodiment, the package includes an initial user sample set and an initial test sample set, each initial user sample in the initial user sample set is a favorite package of each user, each initial test sample in the initial test sample set is a candidate package to be recommended to each user, and the candidate package is a package currently on sale in the operator; the package similarity calculation submodule may include:

[0258] A labeling unit is used to standardize each eigenvalue corresponding to each initial user sample in the package feature matrix of the same user group to obtain a standardized target user sample set, and to standardize each eigenvalue corresponding to each initial test sample in the package feature matrix of the same user group to obtain a labeled target test sample set;

[0259] The package similarity calculation unit is used to calculate the distance between each target user sample in the target user sample set and each target test sample in the target test sample set through the Euclidean distance formula, and obtain the package similarity between each preferred package and each candidate package applicable to the same user group.

[0260] In one embodiment, the recommended package fusion module 340 may include:

[0261] The recommended package selection submodule is used to select the first N recommended packages from the first recommended package set to form a first recommended package list, and select the first N recommended packages from the second recommended package set to form a second recommended package list; wherein N is a positive integer;

[0262] A list merging submodule, used to merge and remove duplicates of the first recommended package list and the second recommended package list belonging to the same user to obtain a third recommended package list;

[0263] The package sorting submodule is used to sort the multiple recommended packages in the third recommended package list of each user according to the launch time to obtain the target recommended package list of each user.

[0264] As can be seen from the example, the scheme provided by the present application obtains the user data of each user and the package data of each package; wherein the user data includes the user historical behavior data and the user demand data; the user historical behavior data and the user demand data are used to calculate the user similarity between two users, and the package data is used to calculate the package similarity between two packages; according to the user similarity, the first recommended package set of each user is determined, and according to the package similarity, the second recommended package set of each user is determined; the first recommended package set and the second recommended package set are integrated to obtain the target recommended package list of each user. The present application can solve the problem that the simple data description cannot allow users to quickly judge whether the package meets their needs by collecting user demand data in real time, and the recommendation based on the user's collaborative filtering algorithm can solve the problem of data sparsity, making the recommendation more accurate, and the recommendation based on the content recommendation algorithm can find potential target user groups for new packages in a short time, thereby solving the package cold start problem, and by combining user-based collaborative filtering and content-based recommendation, a hybrid recommendation strategy is formed, so that not only the similarity between users can be used for recommendation, but also the characteristics of the package itself can be used for recommendation, thereby achieving more accurate personalized recommendation and improving the handling efficiency of marketing acceptance staff.

[0265] Regarding the device in the above embodiment, the specific manner in which each module performs the operation has been described in detail in the embodiment of the method, and will not be elaborated again here.

[0266] Figure 4 It is a schematic diagram of the structure of an electronic device shown in an embodiment of the present application.

[0267] See also Figure 4 , the electronic device 400 includes a memory 410 and a processor 420.

[0268] The processor 420 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or the processor may be any conventional processor, etc.

[0269] The memory 410 may include various types of storage units, such as system memory, read-only memory (ROM), and permanent storage. Among them, ROM can store static data or instructions required by the processor 420 or other modules of the computer. The permanent storage device may be a readable and writable storage device. The permanent storage device may be a non-volatile storage device that does not lose the stored instructions and data even after the computer is powered off. In some embodiments, the permanent storage device uses a large-capacity storage device (such as a magnetic or optical disk, flash memory) as a permanent storage device. In some other embodiments, the permanent storage device may be a removable storage device (such as a floppy disk, optical drive). The system memory may be a readable and writable storage device or a volatile readable and writable storage device, such as a dynamic random access memory. The system memory may store some or all instructions and data required by the processor at run time. In addition, the memory 410 may include any combination of computer-readable storage media, including various types of semiconductor storage chips (such as DRAM, SRAM, SDRAM, flash memory, programmable read-only memory), and disks and / or optical disks may also be used. In some embodiments, the memory 410 may include a readable and / or writable removable storage device, such as a laser disc (CD), a read-only digital versatile disc (such as a DVD-ROM, a double-layer DVD-ROM), a read-only Blu-ray disc, an ultra-density optical disc, a flash memory card (such as an SD card, a mini SD card, a Micro-SD card, etc.), a magnetic floppy disk, etc. The computer-readable storage medium does not include carrier waves and transient electronic signals transmitted wirelessly or wired.

[0270] The memory 410 stores executable codes, and when the executable codes are processed by the processor 420 , the processor 420 can execute part or all of the methods described above.

[0271] In addition, the method according to the present application may also be implemented as a computer program or a computer program product, which includes computer program code instructions for executing some or all of the steps in the above method of the present application.

[0272] Alternatively, the present application can also be implemented as a computer-readable storage medium (or non-transitory machine-readable storage medium or machine-readable storage medium) on which executable code (or computer program or computer instruction code) is stored. When the executable code (or computer program or computer instruction code) is executed by a processor of an electronic device (or server, etc.), the processor executes part or all of the steps of the above-mentioned method according to the present application.

[0273] The embodiments of the present application have been described above, and the above description is exemplary, not exhaustive, and is not limited to the disclosed embodiments. Many modifications and changes will be apparent to those of ordinary skill in the art without departing from the scope and spirit of the described embodiments. The selection of terms used herein is intended to best explain the principles of the embodiments, practical applications, or improvements to the technology in the market, or to enable other persons of ordinary skill in the art to understand the embodiments disclosed herein.

Claims

1. A method for recommending a package, characterized in that: The method comprises: Acquire user data of each user and package data of each package; wherein the user data includes user historical behavior data and user demand data; Using the user historical behavior data and the user demand data, calculate the user similarity between two of the users, and using the package data, calculate the package similarity between two of the packages; Determining a first recommended package set for each of the users according to the user similarities, and determining a second recommended package set for each of the users according to the package similarities; The first recommended package set and the second recommended package set are integrated to obtain a target recommended package list for each of the users.

2. The method according to claim 1, characterized in that The obtaining of user data of each user includes: Configure multiple business scenarios; the multiple business scenarios include scenarios recommended for home users, scenarios recommended for business users, and scenarios recommended for gaming users; Determining a target business scenario to which each user belongs from the multiple business scenarios; Obtaining a recommendation factor of each of the users in the target service scenario, and obtaining usage information of an existing number of each of the users in an operator; The recommendation factor is used as user demand data, and the usage information is used as user historical behavior data to obtain user data.

3. The method according to claim 2, characterized in that The user historical behavior data includes user portrait information; the user historical behavior data and the user demand data are used to calculate the user similarity between two users, including: Using historical user behavior data corresponding to user groups belonging to the same target business scenario, constructing an initial user behavior matrix for each of the user groups; and, Using user portrait information corresponding to user groups belonging to the same target business scenario, constructing an initial user portrait matrix for each user group; and, Using user demand data corresponding to user groups belonging to the same target business scenario, construct an initial user demand matrix for each user group; Based on the initial user behavior matrix, the initial user portrait matrix and the initial user demand matrix of the same user group, the user similarity between any two users in the same user group is calculated.

4. The method according to claim 1, characterized in that: The package data includes applicable user groups; the using the package data to calculate the package similarity between the packages comprises: Using the package data of the same user group, constructing a package feature matrix for each user group; Based on the package feature matrix of the same user group, the package similarity between any two packages applicable to the same user group is calculated.

5. The method according to claim 3, characterized in that: The calculating the user similarity between two users in the same user group based on the initial user behavior matrix, the initial user portrait matrix and the initial user demand matrix of the same user group includes: The initial user behavior matrix, the initial user portrait matrix and the initial user demand matrix of the same user group are respectively configured with corresponding first weight values; wherein each eigenvalue in the initial user demand matrix has a corresponding second weight value; performing weighted processing on each eigenvalue in the initial user behavior matrix according to the corresponding first weight value to obtain a target user behavior matrix of the same user group; and According to the corresponding first weight value, weighting is performed on each eigenvalue in the initial user portrait matrix to obtain a target user portrait matrix of the same user group; and Performing weighted processing on each eigenvalue in the initial user demand matrix according to the corresponding first weight value and the corresponding second weight value to obtain a target user demand matrix of the same user group; The target user behavior matrix, the target user portrait matrix and the target user demand matrix are combined to obtain a user information matrix of the same user group; The user similarity between any two users in the same user group is calculated based on each eigenvalue in the user information matrix of the same user group through a cosine similarity model.

6. The method according to claim 4, characterized in that The packages include an initial user sample set and an initial test sample set, each initial user sample in the initial user sample set is a favorite package of each user, each initial test sample in the initial test sample set is a candidate package to be recommended to each user, and the candidate package is a package currently on sale by an operator; The calculating the package similarity between two packages applicable to the same user group based on the package feature matrix of the same user group includes: Standardizing each eigenvalue corresponding to each of the initial user samples in the package feature matrix of the same user group to obtain a standardized target user sample set, and standardizing each eigenvalue corresponding to each of the initial test samples in the package feature matrix of the same user group to obtain a labeled target test sample set; The distance between each target user sample in the target user sample set and each target test sample in the target test sample set is calculated by the Euclidean distance formula to obtain the package similarity between each preferred package and each candidate package applicable to the same user group.

7. The method according to claim 1, characterized in that The fusing the first recommended package set and the second recommended package set to obtain a target recommended package list for each user includes: Selecting the first N recommended packages from the first recommended package set to form a first recommended package list, and selecting the first N recommended packages from the second recommended package set to form a second recommended package list; wherein N is a positive integer; Merging and removing duplicates of the first recommended package list and the second recommended package list belonging to the same user to obtain a third recommended package list; The plurality of recommended packages in the third recommended package list of each user are sorted according to the launch time to obtain a target recommended package list of each user.

8. A meal recommendation device, characterized in that: The device comprises: A data acquisition module, used to acquire user data of each user and package data of each package; wherein the user data includes user historical behavior data and user demand data; A similarity calculation module, used to calculate the user similarity between two users by using the user historical behavior data and the user demand data, and to calculate the package similarity between two packages by using the package data; A recommended package determination module, configured to determine a first recommended package set for each of the users according to the user similarity, and to determine a second recommended package set for each of the users according to the package similarity; The recommended package fusion module is used to fuse the first recommended package set and the second recommended package set to obtain a target recommended package list for each user.

9. An electronic device, characterized in that: include: processor; as well as A memory having executable codes stored thereon, which, when executed by the processor, causes the processor to execute the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having executable codes stored thereon, which, when executed by a processor of an electronic device, causes the processor to execute the method according to any one of claims 1 to 7.

Images