TWI843662B - A similar user extension system, method and non-volatile computer-readable storage medium thereof based on target user group - Google Patents

Abstract

The present invention provides a similar user extension system, method and non-volatile computer-readable storage medium thereof based on target user group, including a grouping module and a comparison module, wherein the grouping module uses the graphical representation of the SPO (Subject-Predicate-Object) triplet to transform the target user group into the first knowledge graph to group a plurality of first target subgroups, and then the comparison module uses the graphical representation of the SPO triples to transform the first target subgroups and candidate user groups into a second knowledge graph, so as to group a plurality of second target subgroups. Therefore, the present invention provides the user can promote the products they are interested in according to the preferences of each second target subgroup, and improve the expansion efficiency and scope.

Description

A similar user expansion system based on a target user group, a method and a non-volatile computer-readable storage medium thereof

The present invention relates to a similar user expansion technology, and more particularly to a similar user expansion system, method and non-volatile computer-readable storage medium based on a target user group.

In order to enhance brand influence and promote product sales, existing advertising platforms use the Lookalike Audience method, which is to group existing users with the same consumption or browsing behavior into a group, analyze the profile of the group of users, and then find similar user groups based on the profile to push advertisements. The profile can be age, gender, interests, and behavior patterns.

The similar user groups found by the above method are the target user groups, which are the targets of marketing activities and advertising push, and are also the consumer groups most likely to purchase goods or services. Therefore, the more target user groups found, the wider the user range of marketing or advertising push, and the better the results will be.

However, existing methods such as Lookalike Audience often find inappropriate target users due to analytical errors, such as sampling errors and model errors, which affects the effectiveness of marketing or promotion. For example, user profile analysis shows that 80% of men over 35 will buy high-end home appliances, so ads are pushed to 1,000 men over 35, but due to a 20% analytical error, as many as 200 men actually did not buy high-end home appliances, and the promotion effect is limited.

Therefore, how to filter out inappropriate target users based on the existing target user groups, so as to more effectively find more similar potential users, thereby expanding the target user groups and improving the effectiveness of marketing or promotion has become an important topic for technical personnel in this field.

In order to solve the above problems, the present invention provides a similar user expansion system based on a target user group, which includes: a grouping module, which collects multiple preference items and multiple preference values of each target user in a target user group, and calculates the corresponding comprehensive preference value according to the multiple preference values of each target user, so as to generate a first knowledge graph using the multiple preference items and the comprehensive preference values of each target user, so as to be divided into a plurality of first target subgroups according to the first knowledge graph; and a comparison module, which is connected to the grouping module by communication or electrical connection, so as to receive the plurality of first target subgroups, so as to be divided into a plurality of first target subgroups according to the plurality of first target subgroups. A target subgroup selects a candidate user group from a non-target user group, and then calculates the corresponding comprehensive preference value according to the multiple preference values of each candidate user in the candidate user group, so as to generate a second knowledge graph using the multiple first target subgroups and the multiple preference items and comprehensive preference values of each candidate user, wherein the comparison module calculates the intra-group similarity of each first target subgroup according to the second knowledge graph, and calculates the similarity between each target user and each candidate user in each first target subgroup, thereby performing grouping to form a plurality of expanded second target subgroups.

The present invention further provides a similar user expansion method based on a target user group, comprising: a grouping module collects a plurality of preference items and a plurality of preference values of each target user in a target user group, and calculates a corresponding comprehensive preference value according to the plurality of preference values of each target user; the grouping module generates a first knowledge graph using the plurality of preference items and the comprehensive preference values of each target user, and divides the target user into a plurality of first target subgroups according to the first knowledge graph; a comparison module selects a plurality of first target subgroups from a non-target user group according to the plurality of first target subgroups; A candidate user group is selected, and then the corresponding comprehensive preference value is calculated according to the multiple preference values of each candidate user in the candidate user group; the comparison module generates a second knowledge graph using the multiple first target subgroups and the multiple preference items and comprehensive preference values of each candidate user; and the comparison module calculates the intra-group similarity of each first target subgroup according to the second knowledge graph, and calculates the similarity between each target user in each first target subgroup and each candidate user, thereby performing grouping to form a plurality of expanded second target subgroups.

In the aforementioned embodiment, the grouping module normalizes the multiple preference values of each target user and then adds them up and averages them to obtain the comprehensive preference value of each target user; and the matching module normalizes the multiple preference values of each candidate user and then adds them up and averages them to obtain the comprehensive preference value of each candidate user.

In the aforementioned embodiment, the grouping module converts the plurality of preference items and the comprehensive preference values of each target user into the first knowledge graph using a graphical representation of a SPO (Subject-Predicate-Object) triple; and the matching module converts the plurality of first target subgroups and the plurality of preference items and the comprehensive preference values of each candidate user into the second knowledge graph using a graphical representation of the SPO triple.

In the aforementioned embodiment, the comparison module selects a plurality of candidate users who have the same plurality of preference items as the plurality of first target subgroups based on the plurality of preference items of the non-target user group and their plurality of preference degree values, so as to serve as the candidate user group.

In the aforementioned embodiment, the grouping module filters the abnormal groups in the plurality of first target subgroups.

The aforementioned embodiment further includes a verification module, which is in communication or electrical connection with the comparison module to receive the second target subgroup and generate a target user interest scale and a similar user interest scale for each second target subgroup, wherein if the target user interest scale and the similar user interest scale of one of the plurality of second target subgroups are similar, the verification module includes the plurality of similar users of one of the plurality of second target subgroups in the final expanded range; otherwise, if they are not similar, the verification module excludes one of the plurality of second target subgroups.

From the above, it can be seen that the similar user expansion system, method and non-volatile computer-readable storage medium based on the target user group of the present invention converts the target user group into a first knowledge graph by using the graphical representation of SPO triples through the clustering module to cluster multiple first target subgroups, and then the matching module converts the multiple first target subgroups and the candidate user group into a second knowledge graph by using the graphical representation of SPO triples, thereby clustering multiple second target subgroups through the similarity calculation method, so that users can promote products of interest to each second target subgroup according to their preferences, so as to improve the expansion efficiency and expansion scope.

1: Similar user expansion system based on target user group

11: Grouping module

12: Comparison module

13: Verification module

S21A to S25A: Steps

S21B to S25B: Steps

S21C to S22C: Steps

S31 to S33: Steps

Figure 1 is a schematic diagram of the architecture of the similar user expansion system based on the target user group of the present invention.

Figure 2A is a schematic diagram of the clustering method process.

Figure 2B is a schematic diagram of the process of the user similarity comparison method.

Figure 2C is a schematic diagram of the verification method process.

Figure 3 is a schematic diagram of the process of the similar user expansion method based on the target user group of the present invention.

Figure 4A and Figure 4B are schematic diagrams of the first knowledge spectrum.

Figure 5 is a schematic diagram of multiple first target subgroups.

Figure 6 is a schematic diagram of similarity calculation.

Figure 7A is a schematic diagram of the interest scale of the target user.

Figure 7B-1, Figure 7B-2 and Figure 7B-3 are schematic diagrams of interest scales for multiple second subgroups.

The following is a specific and concrete example to illustrate the implementation of the present invention. People familiar with this technology can easily understand other advantages and effects of the present invention from the content disclosed in this manual.

It should be noted that the structures, proportions, sizes, etc. depicted in the drawings attached to this manual are only used to match the contents disclosed in the manual for people familiar with this technology to understand and read, and are not used to limit the conditions under which the present invention can be implemented. Therefore, they have no substantial technical significance. Any modification of the structure, change of the proportion relationship or adjustment of the size should still fall within the scope of the technical content disclosed by the present invention without affecting the effects and purposes that can be achieved by the present invention. At the same time, the terms such as "one", "first", "second", "upper" and "lower" used in this specification are only used to facilitate the clarity of the description, and are not used to limit the scope of the implementation of the present invention. The changes or adjustments in their relative relationships shall be regarded as the scope of the implementation of the present invention without substantially changing the technical content.

FIG1 is a schematic diagram of the structure of the similar user expansion system based on the target user group of the present invention. As shown in FIG1 , the similar user expansion system 1 based on the target user group includes: a grouping module 11, a comparison module 12 and a verification module 13.

Specifically, the similar user expansion system 1 based on the target user group can be established in the same (or different) servers (such as general-purpose servers, file servers, storage unit servers, etc.) and computers and other electronic devices with appropriate computing mechanisms, wherein each module in the similar user expansion system 1 based on the target user group (such as the grouping module 11, the matching module 12 and the verification module 13) can be software, hardware or firmware; if it is hardware, it can be a processing unit, processor, computer or server with data processing and computing capabilities; if it is software or firmware, it can include instructions that can be executed by a processing unit, processor, computer or server, and can be installed on the same hardware device or distributed on different multiple hardware devices.

First, the grouping module 11 automatically groups a target user group. As shown in FIG2A , it is a grouping method flow executed by the grouping module 11, which includes the following steps S21A to S25A:

In step S21A, the grouping module 11 collects multiple preference items and multiple preference values of each target user in a target user group in a specified time period (such as a week, a month, etc.). For example, the multiple preference items may be videos or video types, etc., and the multiple preference values may be viewing completion or viewing hours, etc.; the multiple preference items may be music or music types, etc., and the multiple preference values may be listening completion or listening hours, etc.; or, the multiple preference items may be products or product types, and the multiple preference values may be the number of purchasers or the number of purchases, etc. Therefore, the present invention can be applied to different scenarios that require the expansion of users according to needs.

In step S22A, the grouping module 11 forms a comprehensive preference value of each target user according to the multiple preference values of each target user in the target user group.

In step S23A, the clustering module 11 converts the multiple preference items and the comprehensive preference value of each target user into a first knowledge graph using a graphical representation of a SPO (Subject-Predicate-Object) triple.

In step S24A, the clustering module 11 uses a clustering algorithm (such as Louvain, or other similar or equivalent clustering methods or analysis techniques) to perform clustering based on the first knowledge graph, thereby automatically distinguishing into a plurality of first target subgroups, wherein the target users in each of the first target subgroups have similar interests.

In step S25A, the grouping module 11 filters the abnormal groups in the plurality of first target subgroups. For example, a group with too few users (such as one group with one person or one group with less than n people, etc., where n is a positive integer) can be regarded as an abnormal group and filtered.

Secondly, the matching module 12 is communicated or electrically connected to the grouping module 11 to receive a plurality of first target subgroups for user similarity matching. As shown in FIG. 2B , it is a similarity matching method flow executed by the matching module 12, which includes the following steps S21B to S25B:

In step S21B, the comparison module 12 collects multiple preference items and multiple preference values of a non-target user group in a specified period of time (such as a week, a month, etc.), so as to screen out multiple candidate users with the same multiple preference items as the target user group (i.e., multiple first target subgroups) from the non-target user group as a candidate user group, thereby controlling the number of users to be compared for similarity through the selection of the candidate user group.

In step S22B, the comparison module 12 forms a comprehensive preference value for each candidate user based on the multiple preference values of each candidate user in the candidate user group.

In step S23B, the comparison module 12 converts the plurality of preference items and the comprehensive preference values of each target user in the plurality of first target subgroups, and the plurality of preference items and the comprehensive preference values of each candidate user in the candidate user group into a second knowledge graph using the graphical representation of the SPO triple.

In step S24B, the comparison module 12 calculates the similarity between the target users in each of the first target subgroups based on the second knowledge graph using a similarity algorithm (such as Jaccard index, or other similar or equivalent similarity calculation methods or analysis techniques), and then calculates the intra-group similarity of each of the first target subgroups.

(0,10))，以作為各該第一目標子群組之群內相似度。是以，各該第一目標子群組皆具會有各自的群內相似度，且透過該超參數α的調校，能調整各該第一目標子群組之群內相似度，進而控制相似用戶的數量。 In one embodiment, the comparison module 12 calculates the similarity between the two target users in each of the first target subgroups based on the second knowledge graph, and then takes the highest similarity between each target user and the user to sum up and average, and finally multiplies it by a hyperparameter α (α

(0,10)) as the intra-group similarity of each first target subgroup. Therefore, each first target subgroup has its own intra-group similarity, and by adjusting the hyperparameter α, the intra-group similarity of each first target subgroup can be adjusted to control the number of similar users.

In step S25B, the comparison module 12 uses each of the first target subgroups as a comparison basis to calculate the similarity between each of the target users and each of the candidate users in each of the first target subgroups according to the second knowledge graph and through the similarity algorithm or calculation method, and the similarity between one of the plurality of candidate users and one of the plurality of target users is greater than the similarity between one of the plurality of target users. When the intra-group similarity of the first target subgroup to which the candidate belongs is determined, one of the plurality of candidate users is used as a similar user of the first target subgroup to which the candidate user belongs, thereby allocating each of the candidate users as a similar user to each of the first target subgroups to form an expanded plurality of second target subgroups, wherein each of the second target subgroups includes a plurality of target users and a plurality of similar users.

In one embodiment, when one of the plurality of candidate users is a similar user belonging to the plurality of first target subgroups at the same time, the matching module 12 groups the one of the plurality of candidate users into the first target subgroup with the highest similarity value.

In one embodiment, when one of the plurality of candidate users cannot be assigned to one of the plurality of first target subgroups, the matching module 12 excludes one of the plurality of candidate users.

Furthermore, the verification module 13 is communicated or electrically connected to the comparison module 12 to receive the second target subgroup for interest scale verification. As shown in FIG2C , it is a verification method flow executed by the verification module 13, which includes the following steps S21C to S22C:

In step S21C, the verification module 13 verifies each of the second target subgroups separately to generate a target user interest scale and a similar user interest scale for each of the second target subgroups.

In step S22C, the verification module 13 verifies whether the target user interest scale and the similar user interest scale of each second target subgroup are similar. If the target user interest scale and the similar user interest scale of one of the plurality of second target subgroups are similar, the verification module 13 includes the plurality of similar users of one of the plurality of second target subgroups in the final expanded range; otherwise, if they are not similar, the verification module 13 excludes one of the plurality of second target subgroups.

In one embodiment, the target user interest scale and the similar user interest scale may be radar charts, and each indicator may be the plurality of preference items, and the value of each indicator may be a plurality of preference level values. Specifically, the target user interest scale and the similar user interest scale visualize the preferences and preference levels of the plurality of target users and the similar users in a specified time period, and display the proportions of different interests in the form of a chart (such as a radar chart or other analytical chart).

FIG3 is a schematic diagram of the process of the similar user expansion method based on the target user group of the present invention. As shown in FIG3, the content of this method is roughly the same as that of the above-mentioned embodiment, so the same parts are not repeated here. Among them, this method includes the following steps S31 to S33:

In step S31, a grouping module 11 automatically groups a target user group to obtain a plurality of first target subgroups.

In step S32, a comparison module 12 performs user similarity comparison to obtain a plurality of similar users similar to the plurality of first target subgroups from the non-target user group, thereby allocating each of the similar users to the corresponding plurality of first target subgroups to form a plurality of second target subgroups.

In step S33, a verification module 13 performs interest scale verification to generate the target user interest scale and similar user interest scale of each second target subgroup, thereby performing verification to exclude the second target subgroup with dissimilar interest scale.

In addition, the present invention also discloses a non-volatile or non-transitory computer-readable storage medium, which is applied to a computing device or computer having a processor (e.g., CPU, GPU, etc.) and/or memory, and stores instructions, and can be used to execute this non-volatile or non-transitory computer-readable storage medium through the processor and/or memory to execute the above-mentioned method and each step when executing this non-volatile or non-transitory computer-readable storage medium.

The following is a specific embodiment of the invention based on the similar user expansion system 1 of the target user group, and the repetitions with the above embodiments will not be repeated.

In this embodiment, the potential viewing users (i.e. the similar users mentioned above) who may watch the movie "The End" are found on the video platform, and the viewing behavior records for a specified period of 1 month are used as the data basis, and the time month of the record is set as 2023/1, and this embodiment includes the following three stages:

The first stage is to automatically group the target user groups:

The target user group is 30 users who have watched the video "Men: Mavericks" in January 2023. The multiple preference items include but are not limited to multiple video-related information such as the video and its video type, and the multiple preference level values include but are not limited to multiple viewing behavior data such as viewing completion and viewing time.

As shown in Table 1 below, a grouping module 11 collects user viewing behavior records of a target user group, which includes multiple preference items (i.e., viewed videos and video types) and multiple preference values (i.e., viewing completion, viewing hours) of each target user in the target user group. Specifically, Table 1 at least includes fields such as target user, month, viewed videos, video types, viewing completion, and viewing hours.

Table 1: Viewing behavior records of target users

[0,1]，即作為各該目標用戶之綜合喜好程度值。 The grouping module 11 forms a comprehensive preference value for each target user based on the multiple preference values of the viewing completion degree and viewing time of each target user. Specifically, different data often have different units and different value ranges. For example, the viewing completion is 0~100%, and the viewing time is n hours (n≧0). In order to compare or aggregate different data, the data needs to be normalized. The clustering module 11 will scale the viewing completion and viewing time of each target user to the interval [0,1] by normalization without changing the original distribution. The normalization method includes but is not limited to min-max, Z-score, etc. Finally, the clustering module 11 will add up and average the normalized viewing completion and viewing time (i.e., multiple preference values) of each target user to obtain a single comprehensive data.

[0,1] is the comprehensive preference value of each target user.

For example, taking the video 1 watched by the target user A as an example, the grouping module 11 uses min-max for normalization, where min-max=(original data-minimum value)/(maximum value-minimum value). For this, the min-max value of viewing completion=(24-0)/(100-0)=0.24, where the maximum viewing completion is 100 and the minimum is 0; and the min-max value of viewing time=(3-0)/(10-0)=0.3, where the maximum viewing time is 10 (assuming it is user C) with the most viewing time among all users, and the minimum is 0. Finally, the target user A's comprehensive preference value is obtained by adding the average value=(0.24+0.3)/2=0.27. In this way, the comprehensive preference values of the target user A and other users are added, as shown in Table 2 below:

Table 2: User viewing behavior records (newly added comprehensive preference value)

The clustering module 11 uses the graphical representation of SPO (Subject-Predicate-Object) triples to convert each target user (such as target users A, B, C, etc.) and viewed videos (such as videos 1, 2, 3) in Table 2 into Subject and Object respectively, and then convert the comprehensive preference value and month into the associated form of Predicate; and convert the viewed videos and video types into Subject and Object respectively, and then convert the video type into the associated form of Predicate, and store them in the graph database at the same time, wherein the graphical representation of the SPO triple refers to the associated form representation of (Subject)-[Predicate]-(Object), which is the usual existence form of knowledge graph content. Therefore, as shown in FIG. 4A, it is the first knowledge graph after the target user A is converted, and as shown in FIG. 4B, it is the first knowledge graph after the target user is converted.

Furthermore, the clustering module 11 uses a clustering algorithm (such as Louvain) or a clustering method to cluster the first knowledge graph, thereby automatically distinguishing the user nodes, video nodes, video type nodes and the weights (comprehensive preference values) of the edges on the first knowledge graph into a plurality of first target subgroups (as shown in FIG. 5 ).

Finally, the grouping module 11 filters the abnormal groups in the plurality of first target subgroups, as shown in Table 3 below:

Table 3: Multiple first target subgroups

Therefore, the grouping module 11 regards users G and H in groups with too few users (one group of one person or one group of less than n people, etc.) as target users with relatively large interest differences, and excludes users G and H as abnormal groups, that is, deletes the first target subgroups with group numbers 4 and 5 in the plurality of first target subgroups.

The second stage is user similarity comparison:

A comparison module 12 finds 1,000 users from a non-target user group who watched the same video or the same type of video as the target user group in 2023/1, i.e., users with multiple preferences, as candidate user groups. The number of candidate users in the candidate user group can be adjusted considering the performance of similarity comparison.

Thus, the comparison module 12 obtains the user viewing behavior record of the candidate user group, which includes the multiple preference items (i.e., viewing videos and video types) and multiple preference values (i.e., viewing completion, viewing hours) of each candidate user in the candidate user group. Furthermore, the comparison module 12 is executed in the same regularized manner as the grouping module 11, and similarly, based on the multiple preference items and the comprehensive preference values of each target user in the multiple first target subgroups, and the multiple preference items and the comprehensive preference values of each candidate user in the candidate user group, the SPO triplet is used to convert the multiple preference items and the comprehensive preference values into a second knowledge graph (not shown in the figure).

The comparison module 12 calculates the similarity between the target users in each of the first target subgroups based on the second knowledge graph using a similarity algorithm (such as Jaccard index) or a similarity calculation method, and further calculates the intra-group similarity of each of the first target subgroups.

Specifically, as shown in FIG6 , based on the videos, video types and weights (i.e., comprehensive preference values) of two user nodes adjacent to each other on the second knowledge graph, the similarity between all two users on the second knowledge graph is calculated. For example, the similarity between target user A and target user B is calculated as follows:

(0,10))，以作為各該第一目標子群組之群內相似度。是以，各該第一目標子群組皆具會有各自的群內相似度，如下表4所示： The comparison module 12 calculates the similarity between the two target users in each of the first target subgroups according to the second knowledge graph, and then takes the highest similarity between each target user and the user to sum up and average, and finally multiplies it by a hyperparameter α=0.95 (α

(0,10)) as the intra-group similarity of each first target subgroup. Therefore, each first target subgroup has its own intra-group similarity, as shown in Table 4 below:

Table 4: Intra-group similarity of each first target subgroup

The matching module 12 calculates the similarity between each target user and each candidate user in each first target subgroup, and if the similarity between one of the plurality of candidate users and one of the plurality of target users is greater than the intra-group similarity of the first target subgroup to which the one of the plurality of target users belongs, then the one of the plurality of candidate users is a similar user of the first target subgroup to which the one of the plurality of target users belongs. Furthermore, when one of the plurality of candidate users is a similar user belonging to the plurality of first target subgroups at the same time, the matching module 12 groups the one of the plurality of candidate users into the first target subgroup with the highest similarity value, as shown in Table 5 below:

Table 5: Clustering results of candidate users

Therefore, the matching module 12 assigns the candidate users I, J as similar users to each of the first target subgroups 2, 3 to form a plurality of second target subgroups.

Furthermore, the matching module 12 adjusts the intra-group similarity of each second target subgroup through the adjustment of the hyperparameter α, thereby controlling the number of similar users. The statistical results of similar users of each second target subgroup are shown in Table 6 below:

Table 6: Statistics of similar users of each second target subgroup

The third stage is the validation of the interest scale:

The verification module 13 verifies each of the second target subgroups 1, 2, and 3 separately to generate a target user interest scale and a similar user interest scale for each of the second target subgroups. Specifically, the interest scale visualizes the user group's favorite items and preference levels during a specified period of time, and displays the proportions of different interests in the form of a chart (radar chart or other analysis chart).

Therefore, as shown in FIG. 7A , the verification module 13 calculates the viewing ratio of the target user group in the second subgroup 1 between different video types based on the viewing behavior record in 2023/1, and visualizes the viewing ratio into an interest scale, wherein the viewing ratio can be obtained by the viewing completion degree, viewing hours or a combination thereof of the video, and the calculation method of the viewing ratio is not limited here.

Accordingly, as shown in FIG. 7B-1, FIG. 7B-2 and FIG. 7B-3, the verification module 13 generates corresponding target user interest scales (radar graphs) and similar user interest scales (radar graphs) according to the target user groups and similar user groups in the second subgroups 1, 2, and 3, and then the verification module 13 verifies whether the target user interest scales and the similar user interest scales of each of the second target subgroups 1, 2, and 3 are similar. In this regard, it can be seen from FIG. 7B-1, FIG. 7B-2 and FIG. 7B-3 that the second subgroups 1 and 3 are similar groups, and the second subgroup 2 is a dissimilar group, so the second subgroup 2 can be excluded, and customized video recommendations can be made for the second subgroups 1 and 3 respectively in the subsequent.

In summary, the similar user expansion system, method and non-volatile computer-readable storage medium based on the target user group of the present invention uses the SPO (Subject-Predicate-Object) triples graphical representation method to convert the target user group into a first knowledge graph to group multiple first target subgroups, and then uses the SPO triples graphical representation method to convert the multiple first target subgroups and candidate user groups into a second knowledge graph, thereby grouping multiple second target subgroups through a similarity calculation method, so that users can promote products of interest to each second target subgroup according to their preferences, so as to improve the expansion efficiency and expansion scope.

Furthermore, the similar user expansion system, method and non-volatile computer-readable storage medium of the present invention based on the target user group have at least the following technical differences and their effects:

1. Automatic grouping of target user groups: The target user groups are grouped by interest through grouping calculation methods. On the one hand, it replaces the laborious and time-consuming manual grouping. On the other hand, it can select target user groups with similar interests and filter out target users with large interest differences. Through user groups with similar interests, it is easier to select similar users to achieve the purpose of precise expansion.

2. User similarity comparison: Based on the common preferences and degree of preferences among users, similarity is calculated to facilitate the determination of whether the interests of different users are similar, thereby more effectively assigning similar users to corresponding groups as targets for expansion.

3. Interest scale verification: Visualize the common preferences and preference levels of user groups to generate an interest scale. This makes it easier to verify the differences in interests of different user groups through visualization, thereby improving promotion efficiency.

The above implementation forms are only illustrative of the principles and effects of the present invention, and are not intended to limit the present invention. Anyone familiar with this technology can modify and change the above implementation forms without violating the spirit and scope of the present invention. Therefore, the scope of protection of the present invention should be as listed in the scope of the patent application.

S31 to S33: Steps

Claims (13)

A similar user expansion system based on a target user group includes: A grouping module collects multiple preference items and multiple preference values of each target user in a target user group, and calculates the corresponding comprehensive preference value based on the multiple preference values of each target user, so as to generate a first knowledge graph using the multiple preference items and comprehensive preference values of each target user, so as to be divided into multiple first target subgroups according to the first knowledge graph; and A comparison module is communicated or electrically connected to the grouping module to receive the plurality of first target subgroups, so as to screen a candidate user group from a non-target user group based on the plurality of first target subgroups, and then calculate the corresponding comprehensive preference value based on the plurality of preference values of each candidate user in the candidate user group, so as to generate a second knowledge graph using the plurality of first target subgroups and the plurality of preference items and comprehensive preference values of each candidate user, The comparison module calculates the intra-group similarity of each first target subgroup according to the second knowledge graph, and calculates the similarity between each target user and each candidate user in each first target subgroup, thereby performing grouping to form a plurality of expanded second target subgroups. As described in claim 1, the similar user expansion system based on the target user group, wherein the grouping module normalizes the multiple preference values of each target user and then adds them up and averages them to obtain the comprehensive preference value of each target user; and the matching module normalizes the multiple preference values of each candidate user and then adds them up and averages them to obtain the comprehensive preference value of each candidate user. The similar user expansion system based on the target user group as described in claim 1, wherein the grouping module converts the plurality of preference items and the comprehensive preference value of each target user into the first knowledge graph using a graphical representation of a SPO (Subject-Predicate-Object) triple; and the matching module converts the plurality of first target subgroups and the plurality of preference items and the comprehensive preference value of each candidate user into the second knowledge graph using a graphical representation of the SPO triple. As described in claim 1, the similar user expansion system based on the target user group, wherein the comparison module selects a plurality of candidate users having the same plurality of preference items as the plurality of first target subgroups based on the plurality of preference items of the non-target user group and their plurality of preference degree values, so as to serve as the candidate user group. A similar user expansion system based on a target user group as described in claim 1, wherein the grouping module filters abnormal groups in the plurality of first target subgroups. The similar user expansion system based on the target user group as described in claim 1 further includes a verification module, which is communicatively or electrically connected to the comparison module to receive the second target subgroup and generate a target user interest scale and a similar user interest scale for each second target subgroup, wherein if the target user interest scale and the similar user interest scale of one of the plurality of second target subgroups are similar, the verification module includes the plurality of similar users of one of the plurality of second target subgroups in the final expansion range, and if they are not similar, the verification module excludes one of the plurality of second target subgroups. A similar user expansion method based on a target user group includes: A grouping module collects multiple preference items and multiple preference values of each target user in a target user group, and calculates the corresponding comprehensive preference value based on the multiple preference values of each target user; The grouping module generates a first knowledge graph using the target user's multiple preference items and comprehensive preference values, and divides the target user into multiple first target subgroups according to the first knowledge graph; A comparison module selects a candidate user group from a non-target user group based on the plurality of first target subgroups, and then calculates the corresponding comprehensive preference value based on the plurality of preference values of each candidate user in the candidate user group; The comparison module generates a second knowledge graph using the plurality of first target subgroups and the plurality of preference items and comprehensive preference values of each candidate user; and The comparison module calculates the intra-group similarity of each of the first target subgroups based on the second knowledge graph, and calculates the similarity between each of the target users and each of the candidate users in each of the first target subgroups, thereby performing grouping to form a plurality of expanded second target subgroups. The similar user expansion method based on the target user group as described in claim 7 further includes the grouping module normalizing the multiple preference values of each target user and summing up and averaging them to obtain the comprehensive preference value of each target user; and the matching module normalizing the multiple preference values of each candidate user and summing up and averaging them to obtain the comprehensive preference value of each candidate user. The similar user expansion method based on the target user group as described in claim 7 further includes the grouping module converting the plurality of preference items and the comprehensive preference values of each target user into the first knowledge graph using a graphical representation of an SPO triple; and the matching module converting the plurality of first target subgroups and the plurality of preference items and the comprehensive preference values of each candidate user into the second knowledge graph using a graphical representation of the SPO triple. The similar user expansion method based on the target user group as described in claim 7 further includes the comparison module screening out a plurality of candidate users having the same plurality of preference items as the plurality of first target subgroups based on the plurality of preference items and the plurality of preference degree values of the non-target user group, so as to serve as the candidate user group. The similar user expansion method based on the target user group as described in claim 7 further includes filtering the abnormal groups in the plurality of first target subgroups by the grouping module. The similar user expansion method based on the target user group as described in claim 7 further includes generating a target user interest scale and a similar user interest scale of each second target subgroup by a verification module, wherein if the target user interest scale and the similar user interest scale of one of the plurality of second target subgroups are similar, the verification module includes the plurality of similar users of one of the plurality of second target subgroups in the final expansion range, and if they are not similar, the verification module excludes one of the plurality of second target subgroups. A non-volatile computer-readable storage medium, used in a computing device or a computer, stores instructions for executing a similar user expansion method based on a target user group as described in any one of claims 7 to 12.

Images