Friend recommendation method and system based on digital business card ecology

Abstract

The invention relates to a friend recommendation method and system based on digital business card ecology, and the method comprises the following steps: taking a mobile phone number of each digital business card user as a unique identifier, obtaining user data, and dividing the user data into an interest label score, an industry label score, a position TA List, a friend list and dynamic data; obtaining an interest coincidence index a1, an industry coincidence index a2, a position coincidence index a3 and a friend coincidence index a4 of the user to be recommended and other users, and setting a dynamic coincidence index P (act) and a time decay factor alpha; and setting weights x1, x2, x3 and x4, calculating friend recommendation indexes by using a friend recommendation index model, sorting the friend recommendation indexes, and constructing a friend recommendation list. Compared with the prior art, the friend recommendation method is higher in friend recommendation precision and efficiency and higher in pertinence.

Description

Technical Field

[0001] The present invention relates to the field of big data and social network, and in particular to a friend recommendation method and system based on digital business card ecology. Background Art

[0002] Digital business cards are virtual business cards that can be stored in electronic devices (such as mobile phones or computers) after the data related to paper business cards are digitized. The content of digital business cards often includes name, company unit (affiliated organization) and contact information. With the development of e-commerce and the Internet, digital business cards are easy to store, manage and share, avoiding the cumbersome production and easy loss of paper business cards, and improving the efficiency and convenience of business social interaction. At present, products related to digital business cards include CamCard, AI Business Card, Business Card Delivery, etc.

[0003] Digital business cards carry a lot of information related to business and social networking, and users expect to acquire customers and conduct social activities based on digital business card clients. How to analyze digital business card users and recommend users who meet their business needs in an appropriate way, thereby helping target users reduce customer acquisition costs, has become an urgent problem to be solved. However, the existing technology has not yet provided a mature solution for how to implement friend recommendations based on digital business cards.

[0004] Existing friend recommendation schemes of social software are generally divided into: (1) recommendation based on user address book, (2) recommendation based on user attribute information, (3) recommendation based on user indirect social interaction, and (4) recommendation based on user content information, among which:

[0005] (1) Recommendations based on the user's address book: Get the user's phone number's address book list, and make recommendations if the contact in the address book is also a registered user.

[0006] (2) Recommendation based on attribute information: Matching recommendations are made based on the information entered by registered users, such as companies and schools, and users with the same attributes are recommended to target users. For example, Maimai and Weibo usually present information in the form of "XX University alumni" or "You are from XX company".

[0007] (3) Recommendations based on indirect user social interactions: Using the concept of "second-degree friends", the system recommends users who have close relationships with the user's direct friends and who interact frequently. This type of recommendation can adopt models such as "friends of friends" or "followers of followers". The "friends of friends" strategy is based on the assumption that if a user is interested in the interests of a friend, they are likely to be interested in the information that these friends are interested in. The "followers of followers" strategy is based on the user's interest in the content of the community to which the person they follow belongs, and recommends relevant community content that they may not have directly encountered. Both methods reflect the diversity of user relationships in social networks.

[0008] (4) Recommendations based on user content information: By applying text analysis techniques such as word frequency statistics and semantic analysis, the system deeply analyzes user-generated content, extracts keywords, and combines them with user-defined interest tags. In this way, a vector model that can represent the user's personalized interests is constructed. The system further uses tag mapping and classification systems to identify and recommend other users or content that match the user's interests, presenting them in the form of "Similar interests - XX category", thereby providing users with highly customized recommendations.

[0009] Although these four recommendation methods have their own unique advantages, they also face some limitations. The first two methods based on social connections can effectively alleviate the cold start problem in the recommendation system to a certain extent. However, their granularity in user segmentation is relatively coarse, and they fail to provide sufficient targeting and may not meet users' needs for personalized recommendations.

[0010] In addition, the recommendation method based on user content information mainly relies on the content and tag information of the articles published by users, and makes recommendations through keyword extraction and interest tag matching. Although this method is direct, it fails to fully utilize the dynamic relationship data in social networks and fails to match industry attributes. Therefore, the recommendation effect may appear to be relatively simple and lack depth and diversity. Summary of the invention

[0011] The purpose of the present invention is to overcome the defects of the above-mentioned prior art and to provide a friend recommendation method and system based on the digital business card ecosystem, combining more detailed user behavior analysis, social network structure, industry attribute mining, and real location and real-time interaction data to provide more accurate and personalized recommendation services.

[0012] The purpose of the present invention can be achieved by the following technical solutions:

[0013] A friend recommendation method based on digital business card ecology includes the following steps:

[0014] Using the mobile phone number of each digital business card user as a unique identifier, obtaining user data, and dividing the user data into interest tag scores, industry tag scores, location TAList, friend list and dynamic data;

[0015] Obtain the interest overlap index a1, industry overlap index a2, location overlap index a3 and friend overlap index a4 between the recommended user and other users, and set the dynamic overlap index P(act) and time decay factor ɑ;

[0016] Weights x1, x2, x3 and x4 are set, the friend recommendation index is calculated using the friend recommendation index model, the friend recommendation index is sorted, and a friend recommendation list is constructed.

[0017] Furthermore, the steps for calculating the interest tag score are as follows:

[0018] Get the web pages that the user has recently browsed, segment the obtained web page content, analyze the word frequency categories through dictionary query, and calculate the text interest tag score set Tab1 for each user i as follows;

[0019] Tab1 i ={(Clu11,a1%),(Clu12,a2%),(Clu13,a3%),……,(Clu1 n ,a n %)}

[0020] In the formula, i represents the i-th user, n represents the total number of interest tags, Clu1 i represents the text interest tag, a i % represents the percentage score corresponding to the text interest tag, (Clu1 i ,a i %) represents the text interest tag score;

[0021] Get the media content that the user has recently browsed, use unsupervised machine learning to perform cluster analysis, and calculate the media interest tag score set Tab2 for each user i as follows:

[0022] Tab2 i ={(Clu21,a1%),(Clu22,a2%),(Clu23,a3%),……,(Clu2 n ,a n %)}

[0023] In the formula, i represents the i-th user, n represents the total number of interest tags, Clu2 i Represents a media interest tag, a i % represents the percentage score corresponding to the media interest tag, (Clu2 i ,a i %) represents the media interest tag score;

[0024] The text interest tag score Tab1 i and media interest tag score Tab2 i Perform multimodal fusion to obtain the user interest tag score set Tab3 for each user i as follows:

[0025] Tab3 i ={(Clu31,a1%),(Clu32,a2%),(Clu33,a3%),……,(Clu3 n ,a n %)}

[0026] In the formula, i represents the i-th user, n represents the total number of interest tags, and Clu3 i Represents the user interest tag after multimodal fusion, a i % represents the percentage score corresponding to the interest tag, (Clu3 i ,a i %) represents the interest tag score.

[0027] Furthermore, the cluster analysis includes multiple clustering algorithms such as K-means, hierarchical clustering or DBSCAN.

[0028] Furthermore, the industry label score set Tab4 i as follows:

[0029] Tab4 i ={(Ind1,a1%),(Ind2,a2%),(Ind3,a3%),……,(Ind n ,a n %)}

[0030] In the formula, i represents the i-th user, n represents the total number of industry tags, and Ind i Represents industry label, a i % is the percentage score corresponding to the industry label, (Ind i ,a i %) represents the interest tag score.

[0031] Furthermore, the interest overlap index a1 is the overlap value of the interest tag scores between each user, and is calculated as follows:

[0032]

[0033] In the formula, a i % represents the percentage score corresponding to the interest tag, and the interest tag score set Tab3 is calculated. i Each user's interest tag Clu3 i The corresponding percentage scores are added together, and finally a1 should satisfy 0≤a1≤100%;

[0034] The industry overlap index a2 is the overlap value of the industry label scores between each user, and is calculated as follows:

[0035]

[0036] In the formula, a i % represents the percentage score corresponding to the interest tag, and calculates the industry tag score set Tab4 i The industry label Ind for each user i The corresponding percentage scores are added together, and finally a2 should satisfy 0≤a2≤100%;

[0037] The position coincidence index a3 is a tracking area list TAL of the current base station location of each user. i The percentage of the number of overlapping TAs to the number of its own TAs is calculated as follows:

[0038]

[0039] In the formula, TAL i The tracking area list of the base station positioning where the user's registered mobile phone number is currently located, TAL other The tracking area list of the base station where the mobile phone numbers registered by other users are currently located. Finally, a3 should satisfy 0≤a3≤100%;

[0040] The friend overlap index a4 is the friend list FRI between each user. i The percentage of the number of overlapping parts to the number of one's own friends is calculated as follows:

[0041]

[0042] Where FRI i For the friend list, FRI other For the friend lists of the remaining users, a4 should finally satisfy 0≤a4≤100%.

[0043] Furthermore, the dynamic coincidence index P(act) includes multiple types of network search terms performed by the user to be recommended on that day and the number with which the user is currently calling.

[0044] Furthermore, the time decay factor ɑ takes a value of 1 on the day and decays to 0 on the 30th day, that is, the dynamic decay factor of the i-th day is After the 31st day, the dynamic data will be deleted.

[0045] Furthermore, the friend recommendation index model is as follows:

[0046] y=x1a1+x2a2+x3a3+x4a4+P(act)ɑ

[0047] Where y is the friend recommendation index.

[0048] Furthermore, based on the user's own behavior of adding friends, the weights x1, x2, x3 and x4 are dynamically adjusted to continuously iterate and calibrate the friend recommendation index model.

[0049] According to another aspect of the present invention, a friend recommendation system based on a digital business card ecosystem is provided, comprising:

[0050] A user data processing module, used to obtain user data and divide the user data into interest tag scores, industry tag scores, location TA List, friend list and dynamic data;

[0051] The overlap index calculation module is used to calculate the interest overlap index a1, industry overlap index a2, location overlap index a3 and friend overlap index a4 according to the interest tag score, industry tag score, location TA List and friend list, and set the dynamic overlap index P(act) and time decay factor ɑ;

[0052] The recommendation index sorting module sets weights x1, x2, x3 and x4, calculates the friend recommendation index using the friend recommendation index model, sorts the friend recommendation index, and constructs a friend recommendation list;

[0053] The model optimization module dynamically adjusts weights x1, x2, x3, and x4 based on the user's own behavior of adding friends, and continuously iterates and calibrates the friend recommendation index model.

[0054] Compared with the prior art, the present invention has the following beneficial effects:

[0055] 1. The present invention is based on real users and analyzes through real data, covering the matching of industry attributes and real location information. The location adopts the TA List based on the operator's base station positioning, which is more real-time and accurate, and can switch between cold and hot starts freely. There is no need to import existing address book friends, and it is more suitable for friend recommendations in business scenarios.

[0056] 2. The present invention improves the accuracy and efficiency of friend recommendations by dynamically triggering recommendations based on the user's real-time call behavior and introducing a time decay factor to re-update the dynamic overlapping scenes after a period of time. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] Figure 1 A flowchart of a friend recommendation method based on a digital business card ecosystem;

[0058] Figure 2 This is a block diagram of a friend recommendation method system based on the digital business card ecosystem. DETAILED DESCRIPTION

[0059] The present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments. This embodiment is implemented based on the technical solution of the present invention, and provides a detailed implementation method and specific operation process, but the protection scope of the present invention is not limited to the following embodiments.

[0060] Example 1

[0061] This embodiment provides a friend recommendation method based on digital business card ecology, such as Figure 1 As shown, the following steps are included:

[0062] S1. Use the mobile phone number of each digital business card user as a unique identifier to obtain user data, and divide the user data into interest tag scores, industry tag scores, location TA List, friend list and dynamic data.

[0063] Use crawlers or cooperate with Internet companies to obtain the web pages that the user's mobile phone number has browsed recently (such as in the past 30 days); segment the content obtained by the crawler; analyze the category (Cluster) to which the word frequency belongs through dictionary query; calculate the text interest tag score set Tab1 for each user i as follows:

[0064] Tab1 i ={(Clu11,a1%),(Clu12,a2%),(Clu13,a3%),……,(Clu1 n ,a n %)}

[0065] In the formula, i represents the i-th user, n represents the total number of interest tags, Clu1 i represents the text interest tag, a i % represents the percentage score corresponding to the text interest tag, (Clu1 i ,a i %) represents the text interest tag score.

[0066] In this example, Tab1 i ={(text category 1, 5%), (text category 2, 7%), ..., (text category n, 2%)}.

[0067] Use crawlers or cooperate with Internet companies to obtain the media content that the user's mobile phone number has browsed recently (such as in the past 30 days), and use unsupervised machine learning to perform cluster analysis. Cluster analysis can use clustering algorithms such as K-means, hierarchical clustering, or DBSCAN to calculate the media interest tag score set Tab2 for each user. i as follows:

[0068] Tab2 i={(Clu21,a1%),(Clu22,a2%),(Clu23,a3%),……,(Clu2 n ,a n %)}

[0069] In the formula, i represents the i-th user, n represents the total number of interest tags, Clu2 i Represents a media interest tag, a i % represents the percentage score corresponding to the media interest tag, (Clu2 i ,a i %) represents the media interest tag score.

[0070] In this embodiment, Tab2 i ={(media category 1, 3%), (media category 2, 13%), ..., (media category n, 5%)}.

[0071] Using weighted average, feature cascade, deep learning model and other methods, text interest and media interest are multimodally fused to obtain the final interest tag score set Tab3 for each user. i as follows:

[0072] Tab3 i ={(Clu31,a1%),(Clu32,a2%),(Clu33,a3%),……,(Clu3 n ,a n %)}

[0073] In the formula, i represents the i-th user, n represents the total number of interest tags, and Clu3 i Represents the user interest tag after multimodal fusion, a i % represents the percentage score corresponding to the interest tag, (Clu3 i ,a i %) represents the interest tag score.

[0074] In this embodiment, Tab3 i ={(category 1, 3%), (category 2, 8%), ..., (category n, 7%)}.

[0075] Get the company name (organization name) filled in by the user when registering, as well as the company-related information published in the business card. Use the method in the industry classification method, device, equipment and storage medium based on big data proposed in Chinese patent CN111027318B to classify the company and the published content according to industry (Industry), and obtain the industry label score set Tab4 for each user i as follows:

[0076] Tab4 i={(Ind1,a1%),(Ind2,a2%),(Ind3,a3%),……,(Ind n ,a n %)}

[0077] In the formula, i represents the i-th user, n represents the total number of industry tags, and Ind i Represents industry label, a i % is the percentage score corresponding to the industry label, (Ind i ,a i %) represents the interest tag score.

[0078] In this embodiment, Tab4 i ={(technology, 63%), (communications, 19%), (entertainment, 18%)}.

[0079] Get the tracking area list of the base station positioning where each user's registered mobile phone number is currently located TAL i (TA List, Tracking Area List) is as follows:

[0080] TAL i ={TA x |x∈{TA total}}

[0081] In the formula, i represents the i-th user, TA total A list of tracking areas representing all base station locations.

[0082] For example: TAL i ={TA1, TA3, TA5}.

[0083] Get the friend list of the user's added friends. The friend uses the mobile phone number as the unique identifier, recorded as FRI i as follows:

[0084] FRI i ={User x |x∈{User total}}

[0085] In the formula, i represents the i-th user, User total Represents the set of all registered users.

[0086] For example: FRI i ={177777777777, 18888888888, 19999999999}.

[0087] S2. Obtain the interest overlap index a1, industry overlap index a2, location overlap index a3 and friend overlap index a4 between the user to be recommended and other users, and set the dynamic overlap index P(act) and time decay factor ɑ.

[0088] Calculate the interest overlap index a1 between each user. The interest overlap index a1 is the overlap value of the interest tag scores between each user. The calculation method is as follows:

[0089]

[0090] In the formula, a i % represents the percentage score corresponding to the interest tag, and the interest tag score set Tab3 is calculated. i Each user's interest tag Clu3 i The corresponding a i % numerical value, take the union and add them, and finally a1 should satisfy 0≤a1≤100%;

[0091] In this embodiment, the interest coincidence index a1 of user 1 and each user's interest tag score is: (user 1 user 2-38%, user 1 user 3-18%, user 1 user 3-30%, ..., user 1 user n-75%);

[0092] The interest overlap index a1 of user 2 and each user's interest tag score is: (user 2 user 1 - 38%, user 2 user 3 - 63%, ..., user 2 user n - 39%).

[0093] Calculate the industry overlap index a2 between each user. The industry overlap index a2 is the overlap value of the industry label scores between each user. The calculation method is as follows:

[0094]

[0095] In the formula, a i % represents the percentage score corresponding to the interest tag, and calculates the industry tag score set Tab4 i The industry label Ind for each user i The corresponding a i % numerical value, take the union and add it, and finally a2 should satisfy 0≤a2≤100%.

[0096] In this embodiment, the industry overlap value a2 of user 1 and each user's industry label score is: (user 1 user 2-48%, user 1 user 3-31%, user 1 user 3-26%, ..., user 1 user n-40%);

[0097] The industry overlap value a2 of user 2 and each user's industry label score is: (user 2 user 1-48%, user 2 user 3-10%, ..., user 2 user n-99%).

[0098] Calculate the position overlap index a3 between each user. The position overlap index a3 is used to calculate the tracking area list TAL of the current base station location between each user. i The percentage of the number of overlapping TAs to the number of its own TAs is calculated as follows:

[0099]

[0100] In the formula, TAL i The tracking area list of the base station positioning where the user's registered mobile phone number is currently located, TAL other The tracking area list of the base station where the mobile phone numbers registered by other users are currently located. Finally, a3 should satisfy 0≤a3≤100%;

[0101] In this embodiment, the position coincidence index a3 of user 1 and each user is: (user 1 user 2-33%, user 1 user 3-25%, user 1 user 3-75%, ..., user 1 user n-20%);

[0102] The position coincidence index a3 of user 2 and each user is: (user 2 user 1-33%, user 2 user 3-75%, ..., user 2 user n-100%).

[0103] Calculate the friend list FRI between each user i The percentage of the number of overlapping parts to the number of one's own friends is used to obtain the friend overlap index a4, which is calculated as follows:

[0104]

[0105] Where FRI i For the friend list, FRI other For the friend lists of the remaining users, a4 should finally satisfy 0≤a4≤100%.

[0106] In this embodiment, the friend overlap index a4 of user 1 and each user is: (user 1 user 2-37%, user 1 user 3-16%, user 1 user 3-8%, ..., user 1 user n-0%);

[0107] The friend overlap index a4 of user 2 and each user is: (user 2 user 1-37%, user 2 user 3-57%, ..., user 2 user n-100%).

[0108] Set the dynamic coincidence index P(act) and the time decay factor ɑ.

[0109] The dynamic overlap index P(act) includes the online search terms that the user has conducted on that day, the number of the current call, etc. Call behavior is not limited to call behavior between terminals, but can be extended to communication behavior between Internet apps, such as WeChat calls, DingTalk calls, etc., as long as dynamic data such as call time, call object, and call location can be obtained.

[0110] In this embodiment, the dynamic overlap index for the day is set to 0.99.

[0111] The time decay factor ɑ takes the value of 1 on the day and decays to 0 on the 30th day, that is, the dynamic decay factor of the i-th day After the 31st day, the dynamic data will be deleted.

[0112] S3. Set weights x1, x2, x3 and x4, calculate the friend recommendation index using the friend recommendation index model, sort the friend recommendation indexes, and construct a friend recommendation list.

[0113] The friend recommendation index model is as follows:

[0114] y=x1a1+x2a2+x3a3+x4a4+P(act)ɑ

[0115] Where y is the friend recommendation index.

[0116] In this embodiment, the weights are set to 0.25, 0.25, 0.25, 0.25, and the friend recommendation index model of the user to be recommended is as follows:

[0117] y=0.25a1+0.25a2+0.25a3+0.25a4+0.99ɑ

[0118] Apply the friend recommendation index model, calculate the friend recommendation index, and arrange them in descending order to obtain the sequence C, [user 1 user 2008-99%, user 1 user 109-97%, ..., user 1 user 8088-1%], build a recommendation list, the first item is the most prioritized recommended friend, and so on.

[0119] For user 1, user 2008 is recommended first, followed by user 109, and so on.

[0120] In another preferred embodiment, it also includes:

[0121] S4. Based on the number of friends added by users, dynamically adjust the weights x1, x2, x3 and x4, and continuously iterate and calibrate the friend recommendation index model.

[0122] If the P(act)ɑ value is the maximum when adding friends, reduce the weights x1, x2, x3 and x4 (recommended 0.01);

[0123] The friend recommendation index model is adjusted to:

[0124] y=0.24a1+0.24a2+0.24a3+0.24a4+0.99ɑ

[0125] If you add a friend again, the a1 value is the maximum value, and the weight x1 is increased (0.01 is recommended), and so on.

[0126] The friend recommendation index model is adjusted to:

[0127] y=0.25a1+0.24a2+0.24a3+0.24a4+0.99ɑ

[0128] After dozens of friends are added, the friend recommendation index model will be iterated to the most suitable one for the current user selection.

[0129] If the new user does not have any basic data, the method in this embodiment is used to perform the first friend recommendation to achieve a cold start. If the user already has 50 friends, the method in this embodiment is used to recommend friends to achieve a hot start. The friend recommendation method proposed in this embodiment can realize free switching between cold and hot starts according to the number of friends of the user.

[0130] Example 2

[0131] This embodiment provides a friend recommendation system based on digital business card ecology. Figure 2 As shown, including:

[0132] A user data processing module, used to obtain user data and divide the user data into interest tag scores, industry tag scores, location TA List, friend list and dynamic data;

[0133] The overlap index calculation module is used to calculate the interest overlap index a1, industry overlap index a2, location overlap index a3 and friend overlap index a4 according to the interest tag score, industry tag score, location TA List and friend list, and set the dynamic overlap index P(act) and time decay factor ɑ;

[0134] The recommendation index sorting module sets weights x1, x2, x3 and x4, calculates the friend recommendation index using the friend recommendation index model, sorts the friend recommendation index, and constructs a friend recommendation list;

[0135] The model optimization module dynamically adjusts weights x1, x2, x3, and x4 based on the user's own behavior of adding friends, and continuously iterates and calibrates the friend recommendation index model.

[0136] The rest is the same as in Example 1.

[0137] The preferred specific embodiments of the present invention are described in detail above. It should be understood that a person skilled in the art can make many modifications and changes based on the concept of the present invention without creative work. Therefore, any technical solution that can be obtained by a person skilled in the art through logical analysis, reasoning or limited experiments based on the concept of the present invention on the basis of the prior art should be within the scope of protection determined by the claims.

Claims

1. A friend recommendation method based on digital business card ecology, characterized in that: The following steps are involved: Using the mobile phone number of each digital business card user as a unique identifier, obtaining user data, and dividing the user data into interest tag scores, industry tag scores, location TA List, friend list, and dynamic data; Obtain the interest overlap index a1, industry overlap index a2, location overlap index a3 and friend overlap index a4 of the recommended user and other users, and set the dynamic overlap index P(act) and time decay factor a; Weights x1, x2, x3 and x4 are set, the friend recommendation index is calculated using the friend recommendation index model, the friend recommendation index is sorted, and a friend recommendation list is constructed.

2. The friend recommendation method based on digital business card ecology according to claim 1, characterized in that: The steps for calculating the interest tag score are as follows: Get the web pages that the user has recently browsed, segment the obtained web page content, analyze the word frequency categories through dictionary query, and calculate the text interest tag score set Tab1 for each user i as follows; <h2 style=";text-align:left;direction:ltr">Tab1<h2 style=";text-align:left;direction:ltr"> i <h2 style=";text-align:left;direction:ltr"> (Clu11, a1)(Clu12, a2)(Clu13, a3)(Clu1<h2 style=";text-align:left;direction:ltr"> n <h2 style=";text-align:left;direction:ltr"> ,a<h2 style=";text-align:left;direction:ltr"> n <h2 style=";text-align:left;direction:ltr"> %)} In the formula, i represents the i-th user, n represents the total number of interest tags, Clu1 i represents the text interest tag, a i % represents the percentage score corresponding to the text interest tag, (Clu1 i , a i %) represents the text interest tag score; Get the media content that the user has recently browsed, use unsupervised machine learning to perform cluster analysis, and calculate the media interest tag score set Tab2 for each user i as follows: <h2 style=";text-align:left;direction:ltr">Tab2<h2 style=";text-align:left;direction:ltr"> i <h2 style=";text-align:left;direction:ltr"> (Clu21, a1)(Clu22, a2)(Clu23, a3)(Clu2<h2 style=";text-align:left;direction:ltr"> n <h2 style=";text-align:left;direction:ltr"> ,a<h2 style=";text-align:left;direction:ltr"> n <h2 style=";text-align:left;direction:ltr"> %)} In the formula, i represents the i-th user, n represents the total number of interest tags, Clu2 i Represents a media interest tag, a i % represents the percentage score corresponding to the media interest tag, (Clu2 i , a i %) represents the media interest tag score; The text interest tag score Tab1 i and media interest tag score Tab2 i Perform multimodal fusion to obtain the user interest tag score set Tab3 for each user i as follows: <h2 style=";text-align:left;direction:ltr">Tab3<h2 style=";text-align:left;direction:ltr"> i <h2 style=";text-align:left;direction:ltr"> (Clu31, a1)(Clu32, a2)(Clu33, a3)(Clu3<h2 style=";text-align:left;direction:ltr"> n <h2 style=";text-align:left;direction:ltr"> ,a<h2 style=";text-align:left;direction:ltr"> n <h2 style=";text-align:left;direction:ltr"> %)} In the formula, i represents the i-th user, n represents the total number of interest tags, and Clu3 i Represents the user interest tag after multimodal fusion, a i % represents the percentage score corresponding to the interest tag, (Clu3 i , a i %) represents the interest tag score.

3. The friend recommendation method based on digital business card ecology according to claim 2 is characterized in that: The cluster analysis includes multiple clustering algorithms such as K-means, hierarchical clustering or DBSCAN.

4. The friend recommendation method based on digital business card ecology according to claim 1, characterized in that: The industry label score set Tab4 i as follows: <h2 style=";text-align:left;direction:ltr">Tab4<h2 style=";text-align:left;direction:ltr"> i <h2 style=";text-align:left;direction:ltr"> (Ind1, a1)(Ind2, a2)(Ind3, a3)(Ind<h2 style=";text-align:left;direction:ltr"> n <h2 style=";text-align:left;direction:ltr"> ,a<h2 style=";text-align:left;direction:ltr"> n <h2 style=";text-align:left;direction:ltr"> %)} In the formula, i represents the i-th user, n represents the total number of industry tags, and Ind i Represents industry label, a i % is the percentage score corresponding to the industry label, (Ind i , ai%) represents the interest tag score.

5. The friend recommendation method based on digital business card ecology according to claim 1, characterized in that: The interest overlap index a1 is the overlap value of the interest tag scores between each user, and is calculated as follows: In the formula, a i % represents the percentage score corresponding to the interest tag, and the interest tag score set Tab3 is calculated. i Each user's interest tag Clu3 i The corresponding percentage scores are added together, and finally a1 should satisfy 0≤a1≤100%; The industry overlap index a2 is the overlap value of the industry label scores between each user, and is calculated as follows: In the formula, a i % represents the percentage score corresponding to the interest tag, and calculates the industry tag score set Tab4 i The industry label Ind for each user i The corresponding percentage scores are added together, and finally a2 should satisfy 0≤a2≤100%; The position coincidence index a3 is a tracking area list TAL of the current base station location of each user. i The percentage of the number of overlapping TAs to the number of its own TAs is calculated as follows: In the formula, TAL i The tracking area list of the base station positioning where the user's registered mobile phone number is currently located, TAL other The tracking area list of the base station where the mobile phone numbers registered by other users are currently located. Finally, a3 should satisfy 0≤a3≤100%; The friend overlap index a4 is the percentage of the number of overlapped parts in each user's friend list FRIi to the number of their own friends, and is calculated as follows: Where FRI i For the friend list, FRI other For the friend lists of the remaining users, a4 should finally satisfy 0≤a4≤100%.

6. The friend recommendation method based on digital business card ecology according to claim 1, characterized in that: The dynamic coincidence index P(act) includes multiple types of network search terms performed by the user to be recommended on that day and the number with which the user is currently calling.

7. The friend recommendation method based on digital business card ecology according to claim 1, characterized in that: The time decay factor a takes a value of 1 on the day and decays to 0 on the 30th day, i.e., the dynamic decay factor of the i-th day After the 31st day, the dynamic data will be deleted.

8. The friend recommendation method based on digital business card ecology according to claim 1, characterized in that: The friend recommendation index model is as follows: y=x1a1+x2a2+x3a3+x4a4+P(act)a Where y is the friend recommendation index.

9. The friend recommendation method based on digital business card ecology according to claim 1, characterized in that: Based on the user's own behavior of adding friends, the weights x1, x2, x3 and x4 are dynamically adjusted, and the friend recommendation index model is continuously iterated and calibrated.

10. A friend recommendation system based on digital business card ecology, characterized in that: include: A user data processing module, used to obtain user data and divide the user data into interest tag scores, industry tag scores, location TA List, friend list and dynamic data; The overlap index calculation module is used to calculate the interest overlap index a1, industry overlap index a2, location overlap index a3 and friend overlap index a4 according to the interest tag score, industry tag score, location TA List and friend list, and set the dynamic overlap index P(act) and time decay factor a; The recommendation index sorting module sets weights x1, x2, x3 and x4, calculates the friend recommendation index using the friend recommendation index model, sorts the friend recommendation index, and constructs a friend recommendation list; The model optimization module dynamically adjusts weights x1, x2, x3, and x4 based on the user's own behavior of adding friends, and continuously iterates and calibrates the friend recommendation index model.

Images