TWI679600B - System and method for characteristics prediction - Google Patents

Abstract

A feature prediction system and a feature prediction method thereof firstly provide a plurality of target groups, which include a target group of a tracking account and an index account, and the index account belongs to a known state of the project characteristics, and then the index account is related to the project characteristics The distribution ratio is used as reference information. Then, the target group is used to define the hypothetical probability of the tracking account for the characteristics of the project according to the reference information, and then each of the tracking groups is compared with each other to obtain the prediction of the characteristics of the project for each tracking account. result. Therefore, the present invention only needs to obtain a small number of index account numbers, and can analyze most of the tracking account's prediction results for the characteristics of the item, so as to facilitate effective use by advertisers, brands, or market research consultants.

Description

Feature prediction system and feature prediction method

The invention relates to a feature prediction system, in particular to a feature prediction system and a feature prediction method capable of improving prediction accuracy.

At present, social networking sites are widely used in the online world, such as Facebook, Youtube, Blog, or other online platforms. They allow viewers to leave messages, like, share, reprint, or other ways to express behavioral instructions.

At present, most viewers or users log in to various conventional social networking sites with user accounts (virtual characters). Because the privacy provisions of social networking site providers protect user data, they cannot know the user's personal data.

However, for advertisers, brands, or market research consultants, it is difficult to analyze the distribution of project characteristics (such as gender characteristics, academic characteristics, regional characteristics, or age characteristics) of these social networking sites from these user accounts. It is difficult for advertisers, brands, or market research consultants to analyze and judge the nature of users of these social networking sites based on their behavioral instructions (such as comments, likes, shares, reprints, or other), and it is difficult to use them effectively ( Such as advertising, etc.).

Furthermore, if advertisers, brands, or market research consultants want to effectively determine the nature of users of these social networking sites, they usually need to purchase all personal account user profile information from the owners of the social networking sites, thereby greatly increasing the cost of analysis.

Therefore, how to overcome the problem of conventional technology is an important issue.

In order to solve the problems of the above-mentioned conventional technology, the present invention discloses a feature prediction system including: a data processing module for providing a plurality of target groups including a tracking account and an index account, wherein the index account is for the project feature system. Belongs to a known state, and the distribution ratio of the indicator account to the project characteristics is used as reference information; and the prediction module is used to define the target group's hypothetical probability of the tracking account for the project characteristics based on the reference information, The tracking result is compared with each of the target groups to obtain the prediction result of the tracking account for the characteristics of the item.

The present invention also provides a feature prediction method, which includes: providing a plurality of target groups, including a target group of tracking account numbers and index account numbers, wherein the index account number is a known state for the project characteristics; and the index account number is for the project characteristics. The distribution ratio is used as reference information; the target group is used to define the hypothetical probability of the tracking account for the characteristics of the project according to the reference information; and the tracking account is interactively compared with each of the target groups to obtain each tracking account. The prediction results of the characteristics of the project.

In the foregoing feature prediction system and method, the target group system refers to a collection of behavior instructions of the tracking account and the indicator account on the same network interface.

In the aforementioned feature prediction system and method, the item feature is gender Characteristics, academic characteristics, regional characteristics or age characteristics.

In the foregoing feature prediction system and method, the tracking account is in an unknown state for the item feature before the interactive comparison, and the tracking account is in a predicted state for the item feature after the interactive comparison, and is based on The prediction status is integrated into the prediction result.

In the foregoing feature prediction system and method, the distribution ratio of the target group to the feature of the item is equal to the distribution ratio of the reference information.

In the foregoing feature prediction system and method, the hypothetical probability that each of the tracking account numbers in the target group for the item feature is based on the distribution ratio of the target group to the item feature.

In the foregoing feature prediction system and method, the prediction module further collates the prediction result of the tracking account number with the index account number and categorizes the distribution ratio of the target group to the item feature as prediction information.

It can be known from the above that in the feature prediction system and the feature prediction method of the present invention, the hypothesis probability of the tracking account for the feature of the project is defined mainly by reference information in the target group, and then the target group is interactively compared with each other. Each tracking account is used to obtain the prediction result of each tracking account for the characteristics of the project. Therefore, compared with the conventional technology, the present invention only needs to obtain a small number of index accounts of known project characteristics, and can analyze most of the unknowns. The tracking result of the project characteristics for the project results of the project characteristics, so the forecast information can be sorted out, which not only greatly reduces the analysis cost, but also helps advertisers or other users (such as brand owners or market research consultants) to conduct effective business activities. (Such as advertising) or other follow-up actions.

1‧‧‧ electronic equipment

10‧‧‧Host

11‧‧‧Display device

Figure 1A is a three-dimensional schematic diagram of an electronic device configured with the feature prediction system of the present invention; Figure 1B is a schematic diagram of the architecture of the feature prediction system of the present invention; and Figure 2 is a schematic flowchart of a feature prediction method of the present invention .

The following describes the implementation of the present invention through specific embodiments. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification.

It should be noted that the structures, proportions, sizes, etc. shown in the drawings in this specification are only used to match the content disclosed in the specification for the understanding and reading of those skilled in the art, and are not intended to limit the implementation of the present invention. The limited conditions are not technically significant. Any modification of the structure, change of the proportional relationship, or adjustment of the size should still fall within the scope of this invention without affecting the effects and goals that can be achieved by the present invention. The technical content disclosed by the invention can be covered. At the same time, the terms such as "upper" and "one" cited in this specification are only for the convenience of description, and are not used to limit the scope of the present invention. Substantially changing the technical content shall be regarded as the scope in which the present invention can be implemented.

FIG. 1A is a schematic perspective view of an electronic device configured with the feature prediction system of the present invention. As shown in FIG. 1A, the electronic device 1 is a computer (such as an industrial computer, a desktop computer, a notebook computer, a tablet computer, or a smart phone, etc.), which has a host 10 and an electrical connection. The display device 11 of the host 10, and the host 10 is configured with the feature prediction system The system can cooperate with network equipment for calculation. As shown in FIG. 1B, the feature prediction system includes a data processing module and a prediction module.

The data processing module provides a target group including a tracking account and an indicator account, wherein the indicator account belongs to a state of a known item characteristic, and the distribution ratio of the indicator account to the item characteristic is used as reference information.

In this embodiment, the data processing module has a collection unit and an integration unit. The collection department can provide multiple indicator account numbers with project characteristics. The integration department provides a target group including a tracking account and an indicator account with at least the threshold number according to the threshold number, and uses the distribution ratio of the indicator account to the characteristics of the project as reference information, wherein the tracking account is unknown Status of item characteristics.

Furthermore, the characteristics of the project are gender characteristics (such as men and women), education characteristics (such as primary school or university), regional characteristics (such as location, country or city), or age characteristics (such as 10 to 20 years old).

In addition, the threshold number is set according to demand, and the higher the set value, the higher the prediction accuracy of the prediction module.

In addition, the target group system refers to a collection of behavior instructions of the tracking account and the indicator account on the same network interface. For example, the target group system is a Facebook fan page or community, a Youtube video, a blog, an electronic mall, or other online platforms, and the behavior instruction is a message action, an option action (such as a like), or a reprint action ( (Like sharing).

The forecasting module is to define the hypothetical probability of the tracking account for the characteristics of the project according to the reference information for the target group. The target group interacts with each of the tracking accounts to obtain the prediction results of the tracking account for the characteristics of the item. Therefore, the tracking account is in a state where the characteristics of the item are unknown before the interaction comparison is performed. After comparison, it belongs to the state of predicting the characteristics of the item.

In this embodiment, the distribution ratio of the target group to the feature of the item is used as attribute information, which is equal to the distribution ratio in the reference information, and the hypothetical probability system is obtained based on the distribution ratio in the attribute information.

Furthermore, the prediction module has an analysis unit and a prediction unit, and the analysis unit is used to obtain the hypothesis probability, and the prediction unit is used to obtain the prediction result.

FIG. 2 is a method for performing feature prediction using the feature prediction system of the present invention.

In step one, first, a plurality of indicator account numbers having item characteristics are provided, and the user sets a threshold quantity in the data processing module.

In this embodiment, the collection department of the data processing module collects 31,578 index accounts (such as Facebook users in Taiwan) from online platforms (such as Facebook, Youtube, IG, LINE, e-mall or Blog, etc.), And the project characteristics of the 31,578 indicator accounts belong to a known state. For example, the Facebook system platform provides relevant information about the indicator accounts (such as personal data) to users (such as partners, brands, market research consultants, or advertisers) or the characteristics of the items that users purchase for the indicator accounts. Data, so the user can obtain the project characteristics of the 31578 indicator accounts, and the collection department is organized into the quantitative information shown in Tables 1 and 2 below:

Furthermore, the collection department may number the 31578 index account numbers, that is, No. 1 to 31578, and randomly divide these index account numbers (31578) into two categories, a verification group and an experiment group, as required. For example, 90% of the 31578 indicator accounts are classified as the experimental group (including 28420 indicator accounts, such as No. 1 to 28420) for related operations of the subsequent prediction mechanism, and the 31578 indicators 10% of the account number is classified as the verification group (including about 3158 index account numbers, such as No. 28420 to 31578) for related operations of the subsequent verification mechanism. It should be noted that the larger the number of experimental groups, the better, so the number of experimental groups can be set according to requirements, and even the verification group can be omitted.

In addition, the threshold number represents the number of the indicator account number is "10", and does not represent a specific 10 indicator account numbers. For example, number at least 10 groups such as number 1 to 10, or number 1, 3, 5, 7, 9, 11, 13, 15, 17, and 19, etc.

In addition, the integration department collects multiple known groups (such as Facebook fan pages or communities) from an online platform (such as Facebook) to aggregate into a group collection, and the known group system includes user account data, It has a tracking account and / Or the indicator account. For example, the integration department collects 2596,000 known groups (such as the fan page or community of Facebook users in Taiwan) from the Facebook system through its Graph API (Application Programming Interface, so-called application programming interface). , There are 18 million user account data in total, but in the subsequent steps, the 18 million user accounts need to first remove the 3158 indicator accounts of the verification group. It should be noted that, in practice, there is no need to set up this verification group, so there is no need to remove any indicator account.

In step two, the integration department will select the required target groups from the known groups according to the threshold number.

In this embodiment, the number of thresholds is 10, so the integration unit is based on the 2596,000 known groups, and compares how many index account numbers are overlapped in each known group to select a stack. Known groups with at least 10 indicator accounts are used as target groups. For example, the integration department selected 23533 target groups, and the target account numbers overlapped by the 23533 target groups are not necessarily the same, and the number of overlaps is not necessarily the same, but it must include at least any of the 28420 target account numbers. 10.

Furthermore, the more the target group that meets the conditions (such as the threshold number), the more accurate the subsequent predictions for the tracking accounts in the target group, that is, the more fan pages are used as the target group, the more accurate the subsequent prediction will be. For example, the total number of target groups superimposed with 10 indicator accounts is greater than the total number of target groups superimposed with 200 indicator accounts, so the accuracy of the prediction operation of the target group superimposed with 10 indicator accounts is higher than the superposition The accuracy of the prediction operation for the target group of 200 indicator accounts. Therefore, the threshold should not be set too much, so as to avoid the total number of the target group being too small.

In addition, the integration department uses the distribution ratio of the indicator account to the characteristics of the project as reference information. Specifically, after selecting 23533 target groups, each target group must contain at least any 10 of the 28420 index account numbers. Therefore, the distribution ratio of the index account numbers in the target group to the project characteristics is sorted into the reference information. For example, suppose one of the target groups contains 300 user account data, which is superimposed with 100 indicator accounts, and the 100 indicator accounts are known to have 40 males and 60 females, so the content of the reference information will show 40% Distribution ratio that belongs to male and 60% belongs to female.

In step three, the analysis module of the prediction module defines the distribution ratio of the target group to the feature of the item according to the reference information as attribute information, that is, the distribution ratio of the attribute information is equal to that in the reference information. Distribution ratio.

，其中，Pr(c|p _k )係為第j個目標群體(粉絲頁或社團)之特徵項目c之分佈比例，u _i (c)係為第i號指標帳號之特徵項目c之相關資料，γ _ij 係為特徵項目之量化資料。 In this embodiment, the definition method of the analysis department uses the following formula:

Among them, Pr (c | p _k ) is the distribution ratio of the characteristic item c of the j-th target group (fan page or community), and u _i (c) is the relevant information of the characteristic item c of the index account number i. , Γ _ij is the quantitative data of the feature item.

For example, following the assumption of step two, the distribution ratio of the reference information shows that 40% belong to men and 60% belong to women. Then the analysis department defines the distribution ratio of all (300) user accounts in the target group as 40% belonging to Male and 60% belong to female, so the attribute information of this target group is included Tolerances show a distribution ratio of 40% belonging to men and 60% belonging to women.

For the same reason, if one of the target groups contains 1,000 user accounts, which is superimposed with 100 indicator accounts, and the 100 indicator accounts have known age characteristics and are organized into reference information, the prediction module defines the target The distribution ratio of the 1,000 user accounts in the group is equal to the distribution ratio of the reference information as the attribute information.

In step 4, the analysis department defines the hypothesis probability for each item of the tracking account in the target group based on the attribute information, and then uses the prediction section to obtain the tracking account obtained by each target group. For the hypothetical probability of project characteristics, perform an interactive comparison between the target groups to obtain the prediction result of the tracking account for the project characteristics.

In this embodiment, it is assumed that one of the target groups includes 300 user accounts (for example, including 100 indicator accounts and 200 tracking accounts), and the reference information (or attribute information) shows that 40% belong to men and 60% belong to women. Then, for the 200 tracking accounts, the hypothetical probability that 40% of the gender characteristics belong to men and 60% belong to women, as shown in the table below.

In addition, since the same tracking account may appear in different target groups among the 23533 target groups, the same tracking account may generate different hypotheses. For example, the hypothesis that the tracking account "abc" in one of the target groups (such as the shopping fan page) has a 40% probability of gender characteristics belongs to Men and 60% belong to women, and this tracking account "abc" in another target group (such as sports fan page) assumes that 80% of the gender characteristics belong to men and 20% belong to women. The content of the hypothetical probability obtained by the account "abc".

At this time, the system needs to perform an interactive comparison to obtain the prediction result of the tracking account for the characteristics of the project. Specifically, the prediction unit is based on a Bayesian framework prediction method to obtain the prediction result of the tracking account for the characteristics of the item, that is, users who perform the same behavior instruction can be classified into users of the same category. The principle is that when the same tracking account has different hypothetical probabilities for the characteristics of the project, the prediction result of the tracking account for the characteristics of the project is determined according to the overlapping number of the target group.

For example, to continue the above list, if the number of superimposed pages of the shopping fan page is 100 (that is, 100 sets of index accounts) and the number of superimposed pages of the sports fan page is 20 (that is, 20 sets of index accounts), the prediction department The hypothetical probability that the shopping fan page is selected (40% belong to men and 60% belong to women) as the prediction result of the tracking account "abc", so that the prediction result of the tracking account "abc" for gender characteristics is female (because of 60% Belongs to women).

It should be understood that if there is only one hypothetical probability for the same tracking account, the prediction department will determine the prediction result of the tracking account for the characteristics of the project after the interactive comparison of each target group according to the single hypothetical probability.

，其中，u _i 係為某一個追蹤帳號，p _j 至p _k 係為包含u _i 的目標群體，P _i 係為一組假設機率的資料(即於所有目標群體p _i,j 中，交互比對包含u _i 的目標群體p _j 至p _k )，Pr(γ|P _i )係為交互比對後的機率資料，Pr(γ|u _i )係為某一個追蹤帳號u _i 之預測結果。 In addition, the prediction module performs the following calculation when performing the interactive comparison: First, the analysis unit sets Pr ( γ | p _{i, j} ) = Pr ( γ | p _j ) to obtain the hypothesis probability, where γ Is the feature of the project, and p _{i, j} is all target groups (such as 23533 fan pages or communities), Pr (γ | p _{i, j} ) is the distribution ratio of reference information, and Pr (γ | p _j ) is Distribution ratio of attribute information of the jth target group (or hypothetical probability of tracking account number). Then, let the prediction department perform an interactive comparison, such as the following calculation process:

, Where u _i is a certain tracking account, p _j to p _k are target groups containing u _i , and P _i is a set of hypothetical data (that is, among all target groups p _{i, j} , the interaction ratio target groups p _j comprises u _i to _{p k), Pr (γ |} P i) system as an interactive than the probability data after the pair, Pr (γ | u _i) based prediction one tracking account u _i of the result.

等於

Specifically, in the above calculation operation, the prediction module is a BigQuery software using Google architecture, which belongs to an enterprise data warehouse and is software written by a SQL (Structured Programming Language) query program. If the prediction result is obtained with super fast speed, but because the feature prediction system only needs to predict, there is no need to write any SUM aggregate function in the SQL. therefore,

equal

However, the user can also use other calculation software to perform the above calculation operations, and there is no particular limitation.

In addition, when the project feature has two ethnic groups, such as male and female groups of gender, one-way calculation can be performed according to the process of steps 1 to 4 above, but when the project feature has more than three ethnic groups, such as age , Address area or education, etc., the same tracking account may have multiple sets of assumptions with the same probability and the same number of overlaps for each target group. Therefore, the absolute prediction of multiple calculations can be performed according to the process of steps 1 to 4 above, that is, "Yes" or "No" needs to be predicted in order to obtain the prediction result of the tracking account. Its contents are presented, for example, 13-17 years old, non 18-24 years old, non 25-34 years old, non 35-44 years old, Non-45-54 and non-55-64 years.

For example, the same tracking account "abc" shows that 60% belong to the age of 13-17 and 60% belong to the age of 18-24, and the overlapping number of the two target groups is 100 index accounts, so the absolute prediction operation through multiple calculations The number of overlaps obtained for 10% belonging to non-13-17 and 80% belonging to non-18-24, and 10% belonging to non-13-17 target groups is 50 index accounts, and 60% The superimposed number of non-18-24 year-old target groups is 100 index accounts, so the prediction module judges that the tracking account "abc" is not 18-24 years old, and therefore predicts that the tracking account "abc" has an age characteristic of 13 -17 years old as the prediction result.

Therefore, the absolute prediction operation of the multiple calculation can improve the accuracy of the prediction result of the tracking account.

It should be noted that if the prediction result of the tracking account for the characteristics of the project cannot be accurately determined through the absolute forecasting operation, the system will define the tracking account as a “fake account”. It should be understood that if the total number of these target groups is enough, the number of "fake account numbers" is very small.

In step 5, since the prediction results of each tracking account have been predicted, the forecasting department can combine the prediction results of each tracking account with the known results of the index account and sort them into the characteristics of the item in each target group. Distribution ratio as prediction information.

For example, in the shopping fan page of 300 user accounts, the gender characteristics of 100 index accounts are known status, and in conjunction with the 200 gender tracking results of the tracking account, the predicted information will show that 25% belong to men and 75% The distribution ratio that belongs to women is shown in the table below.

Similarly, it can be known that according to the prediction methods of steps 1 to 5 above, the distribution ratio of the age characteristics of any target group can be predicted.

In this embodiment, the prediction information ignores the “fake account number” and does not include the distribution ratio. It should be understood that if the number of user accounts in the target group is sufficient (such as more than 100), the impact is not significant because the number of "fake accounts" is very small. For example, if the target group of 1000 user accounts ignores 10 “fake accounts”, the prediction module sorts the distribution ratio of 990 user accounts to the project characteristics into prediction information.

On the other hand, in order to verify the prediction accuracy of the prediction methods of steps 1 to 5 above, a verification mechanism may be used for the verification group. It should be noted that in practice, users do not need to use the authentication mechanism.

In this embodiment, the verification operation is performed on the index accounts (a total of 3158 index accounts) with numbers 28420 to 31578. For example, first select 3,000 indicator accounts from 3158 indicator accounts, assuming (actually known status) that the status of their project characteristics is unknown, as the tracking account (defined below as the verification account), and then the 3000 verification accounts Perform the above-mentioned prediction process of steps 1 to 4 to predict the project characteristics of the 3000 verification account, and then compare its prediction result with its known state, to know the accuracy of the prediction process of steps 1 to 4.

，其中，該準確值(Accuracy)係為3000個驗證帳號經絕對預測作業之預測結果之正確率，該精確值(precision)係為該3000個驗證帳號之預測結果之正確率，該回應值(recall) 係為針對該項目特徵之某一類別(如年齡特徵之13-17歲)之驗證帳號(500個)之預測結果之正確率，整合值(F1)係為該驗證機制之預測正確率，如下兩列表所示： In this embodiment, the verification mechanism uses an integrated method to analyze the accuracy, and the specific formula is as follows:

Among them, the Accuracy is the correct rate of the prediction result of the 3000 verified accounts after absolute prediction operation, the Precision is the correct rate of the prediction result of the 3000 verified accounts, and the response value ( (recall) is the accuracy rate of the prediction results of the verification account (500) for a certain category of the project characteristics (such as the age characteristics of 13-17 years), and the integration value (F1) is the prediction accuracy rate of the verification mechanism , As shown in the following two lists:

Therefore, it can be known from the verification mechanism that after comparing the accurate value and the integrated value, the gender characteristics are only divided into two ethnic groups, so one-way calculation can obtain the correct rate that meets the expectations, and the age characteristics are divided into six ethnic groups. Therefore, it is necessary to perform an absolute prediction operation of multiple calculations to obtain a correct rate (ie, an accurate value) that meets expectations.

In addition, the reasons affecting the predicted information in step 5 above include the "fake account" factor, as well as the behavior instruction, that is, the user account is affected by family members or friends to express behavior instructions on the same network interface; or, It also includes too few user accounts for a certain group, such as the elderly.

In summary, users (such as advertisers, brands, (Analyst, poll analyst, or other users) Utilizing the feature prediction system and feature prediction method of the present invention, it is only necessary to obtain a small number of index accounts of known project features, and then analyze most of the tracking accounts for the project features. The prediction results can be sorted out (for example, inferring the age distribution of the number of "likes" of a fan page to know the public opinion tendency), which not only greatly reduces the analysis cost, but also benefits users (advertisers, brands Business, market research consultants, poll analysts, or other users) can perform effective commercial activities (such as advertising) or other follow-up actions based on these target groups (community websites).

For example, the products of advertisers or brands are women's products, so advertisers or brands can advertise on fan pages or societies where the proportion of women is more than 50%, instead of placing advertisements on each fan page or societies.

The above embodiments are used to exemplify the principle of the present invention and its effects, but not to limit the present invention. Anyone skilled in the art can modify the above embodiments without departing from the spirit and scope of the present invention. Therefore, the scope of protection of the rights of the present invention should be listed in the scope of patent application described later.

Claims (14)

A feature prediction system includes: a data processing module for filtering out a target group including a tracking account and an index account in a superimposed manner, wherein the index account belongs to a known state of a project feature, and the index The distribution ratio of account numbers to the characteristics of the project is used as reference information; and the prediction module is used to define the target group's hypothetical probability of the tracking account for the characteristics of the project based on the reference information, so as to be between the target groups. The tracking account is compared with each other to obtain a prediction result of a single tracking account for the characteristics of the item. The feature prediction system described in item 1 of the scope of patent application, wherein the target group system refers to a collection of behavioral instructions of the tracking account and the indicator account on the same network interface. The feature prediction system described in item 1 of the scope of patent application, wherein the item feature is a gender feature, a schooling feature, a regional feature, or an age feature. The feature prediction system according to item 1 of the scope of patent application, wherein the tracking account is unknown to the feature of the item before the interaction comparison, and the tracking account is about the feature of the item after the interaction comparison. The system belongs to the prediction state, and is integrated into the prediction result according to the prediction state. The feature prediction system described in item 1 of the scope of patent application, wherein the distribution ratio of the target group to the characteristics of the item is equal to the distribution ratio of the reference information. The feature prediction system described in item 1 of the scope of the patent application, wherein the hypothesis probability of each of the tracking account numbers in the target group for the feature of the item is based on the distribution ratio of the target group for the feature of the item. The feature prediction system described in item 1 of the scope of patent application, wherein the prediction module further collates the prediction result of the tracking account number with the indicator account number into the distribution ratio of the target group to the item characteristics as a prediction Information. A feature prediction method includes: sieving a plurality of target groups in a superimposed manner, which includes target groups for tracking account numbers and index account numbers, wherein the index account number is a known state for a project characteristic; and the index account number is for the project. The distribution ratio of the features is used as reference information; the target group is used to define the hypothetical probability of the tracking account for the characteristics of the item according to the reference information; and each tracking group is interactively compared with each of the target groups to obtain a single tracking account The prediction results for the characteristics of the project. The feature prediction method according to item 8 of the scope of patent application, wherein the target group system refers to a collection of behavioral instructions of the tracking account and the indicator account on the same network interface. The feature prediction method according to item 8 of the scope of the patent application, wherein the item feature is a gender feature, a schooling feature, a regional feature, or an age feature. The feature prediction method as described in item 8 of the scope of patent application, wherein the tracking account is in an unknown state for the item feature before the interactive comparison, and the tracking account is for the item feature after the interactive comparison. The system belongs to the prediction state, and is integrated into the prediction result according to the prediction state. The feature prediction method according to item 8 of the scope of patent application, wherein the distribution ratio of the target group to the characteristics of the item is equal to the distribution ratio of the reference information. The feature prediction method described in item 8 of the scope of the patent application, wherein each of the tracking accounts in the target group assumes a probability of the feature of the item based on the distribution ratio of the target group to the feature of the item. The feature prediction method described in item 8 of the scope of patent application, further includes collating the prediction result of the tracking account number with the index account number and sorting the distribution ratio of the target group for the item feature as prediction information.