Server and computer program

Abstract

To provide a technique capable of more efficiently selecting influencers suited to advertisement target products and services.SOLUTION: A server performs the steps of: acquiring vectorized influencer information related to each of a plurality of influencers; acquiring commercial material information related to a commercial material, which is an advertisement target product or service; vectorizing the acquired commercial material information; calculating a similarity score from the vectorized influencer information and the vectorized commercial material information; selecting a first influencer having a high relevance to the commercial material from among the plurality of influencers on the basis of the calculated similarity score; and outputting the first influencer information.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present invention relates to a server and a computer program. [Background technology]

[0002] In recent years, advances in communication technology have led to the provision of social networking services (SNS), which allow a wide range of information to be shared among a large number of users. On SNS, any user can disseminate a variety of information, opinions, assertions, reviews, etc., and among these users, those who disseminate information with great influence are called "influencers."

[0003] Influencers are users who have a high degree of influence on a particular platform, and generally have a large number of followers, subscribers, plays, and views. In recent years, "influencer marketing," which utilizes the influence of influencers to promote a target, has been attracting attention. Patent Document 1 discloses a technology that mediates between advertisers and influencers who are suitable for the product or service being advertised. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] Japanese Patent Application Publication No. 2023-535974 Summary of the Invention [Problem to be solved by the invention]

[0005] The present invention aims to provide a technology for more efficiently selecting influencers suited to products or services to be advertised. [Means for solving the problem]

[0006] According to one aspect of the present invention for solving the above problem, there is provided a server including one or more processors, a memory, and a program stored in the memory, wherein the program, when executed by the one or more processors, causes the server to: obtaining vectorized influencer information for each of a plurality of influencers; Obtaining product information relating to a product that is a product or service being advertised; Vectorizing the product information; Calculating a similarity between the vectorized influencer information and the vectorized product information; selecting a first influencer having a high degree of relevance to the product from among the plurality of influencers based on the calculated similarity; and outputting information about the first influencer. [Effects of the Invention]

[0007] According to the present invention, it is possible to provide a technology for more efficiently selecting influencers suited to products or services to be advertised. [Brief explanation of the drawings]

[0008] [Figure 1] FIG. 1 is a diagram showing an example of the configuration of a system 10 according to an embodiment. [Figure 2] FIG. 1 is a diagram showing an example of a hardware configuration of an information processing apparatus according to an embodiment. [Figure 3] 6 is a flowchart corresponding to an example of processing executed in the server 101 according to the embodiment. [Figure 4] 10 is a flowchart corresponding to another example of the processing executed in the server 101 according to the embodiment. [Figure 5A] FIG. 2 is a diagram showing an example of a display screen of a client terminal 102 according to the embodiment. [Figure 5B] FIG. 10 is a diagram showing another example of the display screen of the client terminal 102 according to the embodiment. [Figure 5C] FIG. 10 is a diagram showing yet another example of the display screen of the client terminal 102 according to the embodiment. [Figure 5D] FIG. 10 is a diagram showing yet another example of the display screen of the client terminal 102 according to the embodiment. [Figure 5E] FIG. 10 is a diagram showing yet another example of the display screen of the client terminal 102 according to the embodiment. DETAILED DESCRIPTION OF THE INVENTION

[0009] Hereinafter, the embodiments will be described in detail with reference to the accompanying drawings. Note that the following embodiments do not limit the scope of the invention as claimed, and not all combinations of features described in the embodiments are necessarily essential to the invention. Two or more of the features described in the embodiments may be arbitrarily combined. Furthermore, the same reference numerals are used for the same or similar components, and redundant explanations will be omitted.

[0010] <System configuration> 1 shows a general configuration of a system 10 corresponding to an embodiment. Here, a server 101 which is an information processing device that implements a service corresponding to this embodiment, a client terminal 102 which is an information processing device used by a client that receives information provided from the server 101, and an SNS server 103 which provides information on influencers to the server 101 are connected via a network 104 such as the Internet. An SNS user information database 105 and a qualitative information database 106 are connected to the server 101, and an SNS information database 107 is connected to the SNS server 103.

[0011] The server 101 is an information processing device that mainly executes processing corresponding to this embodiment. The server 101 collects and acquires information about registered users who are posting information on the SNS server 103 via the network 104 using an API provided by the SNS server 103, and stores the information in the SNS user information database 105. Regarding the information about SNS users to be registered in the SNS user information database 105, only information about SNS users who satisfy predetermined conditions may be registered. The predetermined conditions include, for example, that the number of posts on the channel or account of the SNS user is a predetermined number or more, that the number of follower registrations (also simply referred to as the number of followers) of the channel or the like is a predetermined number or more, that the total number of views of the content of the channel or the like, or the maximum number of views of a single content, is a predetermined number or more, and that the number of high ratings (so-called "likes") for the content of the channel or the like is a predetermined number or more.

[0012] The server 101 also generates summary information and qualitative information as influencer information based on the information of the SNS users stored in the SNS user information database 105, and stores this information in the qualitative information database 106. Methods for generating summary information and qualitative information will be described later. This information may be updated, for example, at regular intervals, or updated as appropriate in response to updates to the registered content of the SNS user information database 105.

[0013] The server 101 can also acquire information on products or services (collectively referred to as "merchandise") that the client wishes to advertise from the client terminal 102, select (or select, the same applies below) SNS users suitable for advertising the merchandise based on the qualitative information stored in the qualitative information database 106, and present a list of SNS users to the client terminal 102. In this embodiment, the type of merchandise that can be advertised is not limited. Any type of product or service can be targeted.

[0014] The client terminal 102 can be configured as any information processing device, such as a general-purpose personal computer or a smartphone. The client terminal 102 transmits information about products or services that the client wishes to advertise to the server 101, and receives from the server 101 a list of SNS users suitable for advertising the products or services. Although only one client terminal 102 is shown in FIG. 1, this is for illustrative purposes only; multiple client terminals 102 can be connected to the network 104, and multiple different clients can receive information from the server 101 via their respective client terminals 102.

[0015] The SNS server 103 is a server that provides a social networking service (SNS) on the network 104. The SNS information database 107 registers information that the SNS server 103 uses to provide the SNS service, and this information includes information about users of the SNS (SNS users). The SNS server 103 provides the server 101 with information about SNS users contained in the SNS information database 107 via an API.

[0016] In this embodiment, a client using the client terminal 102, for example, sells products or services in Japan, and advertises on social media and other media to increase awareness of the products or services in order to promote their sales. The client also aims to further increase awareness by having influencers feature the products or services on their own social media accounts or channels. Since the effectiveness of advertising varies depending on who is asked to do the advertising, the selection of the influencer is important. Therefore, the client requests the server 101 to provide information on influencers suitable for advertising the products or services the client handles. Other possible reasons for selecting an influencer include, for example, the following:

[0017] - To select the destination for placement-based TrueView ads on video streaming sites, which specify the channel to which the ad will be delivered. TrueView ads are ads that are displayed in an interruptive manner while a user is trying to watch a video. - To find influencers who are highly compatible with the target product as interview subjects for a new product monitoring survey. The purpose is to input the PR composition plan (video flow) you are considering and search for influencers that match that composition plan.

[0018] In this embodiment, in response to a request from a client terminal 102, the server 101 provides information on influencers suitable for promoting products and services handled by the client of the client terminal 102. When selecting an influencer, it is desirable to determine the suitability of the product or service and the influencer not only based on indicators such as the size of the influencer's fan base and gender and age, but also taking into consideration multifaceted factors that are difficult to quantify, such as the influencer's worldview, hobbies and preferences, and relationship with fans. In this embodiment, the server 101 selects influencers (first influencers) that are highly suitable for the client's products and services based on qualitative information managed in the qualitative information database 106, and provides the information to the client terminal 102 as an influencer list.

[0019] 1 illustrates the SNS server 103 as a single server, the embodiment is not limited to this, and multiple different SNS servers 103 may be provided depending on the type and platform of the SNS service. Platforms include a video distribution platform, a still image and video distribution platform, a text message distribution platform, a multimedia distribution platform including still images, video, and text, etc. One platform may include multiple servers and databases.

[0020] <Hardware configuration> Next, the hardware configuration of the server 101 will be described with reference to Fig. 2. The server 101 can be configured as an information processing device, for example, from one or more personal computers. In Fig. 2, a CPU / GPU 201 as one or more processors controls the server 101 using programs and data stored in a RAM (random access memory) 202, a ROM (read only memory) 205, an internal storage device 207, and the like as memories, and executes processing corresponding to the embodiments described below. The RAM 202 has an area for reading processing programs stored in the internal storage device 207 and information stored in the external storage device 208, and also has a work area used by the CPU / GPU 201 when executing various processes.

[0021] The input unit 203 is an input means for receiving input from the administrator of the server 101, and is composed of a keyboard, a mouse, etc. The communication I / F (interface) 204 functions as an I / F for connecting to the network 104, etc. The ROM 205 stores programs (such as a boot program) that control the entire server 101, etc. The display unit 206 is a display unit that serves as a display screen, and is composed of a liquid crystal display device, etc.

[0022] The internal storage device 207 is mainly composed of a hard disk, and stores programs and various application data for the server 101 to execute processes. The data stored therein is read out to the RAM 202 as needed. The external storage device 208 is a database, and corresponds to the SNS user information database 105 and the qualitative information database 106, in which various information is stored as needed. The bus 209 provides interconnection between the above-mentioned blocks.

[0023] FIG. 2 has been described as the hardware configuration of the server 101, but the basic hardware configuration of the user terminal 102 and the SNS server 103 can also be the same as that shown in FIG.

[0024] <Generation of qualitative information> The following describes processing in the server 101. Fig. 3 is a flowchart showing an example of processing in the server 101. The processing corresponding to this flowchart can be realized, for example, by the CPU / GPU 201 of the server 101 executing a corresponding program (stored in the ROM 205, the internal storage device 207, etc.).

[0025] In S301, the server 101 uses the API of the SNS server 103 to collect and acquire information about specific registered users of the services provided by the SNS server 103. The acquired information is stored in the SNS user information database 105. Examples of such services include, but are not limited to, YouTube (registered trademark), Instagram (registered trademark), TikTok (registered trademark), Facebook (registered trademark), and X (formerly Twitter (registered trademark)). The registered user information includes, or at least one of, the registrant's profile information, information related to the content posted by the registrant on the SNS (e.g., post title, summary information, hashtags, comments, etc.), and information about the gender and age distribution of viewers (audiences) of the posts. The method of acquiring SNS user information is not limited to the above-described method. The server 101 may also acquire the information by sending or providing a specific questionnaire to the SNS user and receiving the responses from the SNS user. Alternatively, the SNS user may be interviewed and the responses obtained therefrom may be acquired as SNS user information. The SNS users whose information is collected and acquired in S301 may be SNS users who satisfy certain conditions, such as posting to a certain SNS a predetermined number of times or more, having a predetermined number of followers or more, having a predetermined number of total views or more, or having a predetermined number of likes or more. In this embodiment, SNS users who satisfy such conditions are called influencers.

[0026] In the next step S302, the acquired information is integrated to generate summary information. The source information for generating the summary information may be information acquired in any one of the acquisition formats described in relation to S301, or information acquired in any two or more formats. Therefore, summary information can be generated only from information acquired through questionnaire responses or only from information acquired through interview responses. The information collected in S301 is any information including fragmentary keywords and words, and summary information is generated as text information from the collected information. The summary information can be generated using mechanisms such as large language models (LLMs), recurrent neural networks (RNNs), or machine learning. For LLMs, a Transformer-based model (implemented in GPT-4) can be used, for example. For example, summary information integrating acquired information about a certain influencer can be generated as follows:

[0027] Example of summary information: Beauty channel of beauty expert X. X posts videos related to the beauty field, introducing trendy products and products for achieving clear skin. Targeting women aged 18-34, she also offers makeup techniques that can be easily tried at low cost. The channel is easy for viewers to watch, with many short videos presented in a friendly tone.

[0028] In the next step S303, the generated summary information text is vectorized to generate qualitative information. The qualitative information is information representing qualitative characteristics of the influencer, which is used in the server 101 to compare with product or service information. The summary information can be vectorized (embedded representation) by converting the words in the summary information into features that provide an appropriate degree of similarity. In the vectorization process, the text is converted into an array of a predetermined dimension. For example, OpenAI Embeddings (text-embedding-3) converts the text into an array of up to 3072 dimensions. In vectorization, whether the text is long or short, it is converted into a vector of the same number of dimensions. For comparison, the number of dimensions of the vectors must be the same. In this way, the vectorization process can be performed using, for example, OpenAI Embeddings, but is not limited to this.

[0029] In the following S304, the summary information generated in S302 and the qualitative information generated in S303 are stored in the qualitative information database 106. The above process can be performed periodically for each piece of summary information of the influencer.

[0030] <Selection of the first influencer> Next, a method for selecting a first influencer will be described. In this embodiment, based on qualitative information, which is vectorized influencer information stored in the qualitative information database 106, the server 101 selects influencers suitable for advertising products or services that the client wishes to advertise. In this embodiment, the selected influencers are called first influencers. The first influencer may be a single person or multiple people.

[0031] 4 is a flowchart showing an example of processing executed in the server 101. The processing corresponding to this flowchart can be realized, for example, by the CPU / GPU 201 of the server 101 executing a corresponding program (stored in the ROM 205, the internal storage device 207, etc.).

[0032] In S401, the server 101 receives input of information about a commercial product, such as a product or service, from the client terminal 102. In this embodiment, a case where information about beauty-related products is received will be described. The received information can be, for example, a sentence in any format, rather than a list of words such as product name or type. For example, the following sentence can be received:

[0033] Example: A-Room offers beauty products such as skincare bath salts and towels. In 2023, the company released a bath salt containing placenta, hyaluronic acid, honey, and other ingredients, garnering attention for its moisturizing and firming properties. That same year, the company launched a private bath with a soaking serum concept, a receipt campaign, and the sale of the "〇×△" Imabari skincare towels. The brand also conducted several promotional activities, including online sales of assorted sets. The skincare brand "A-Room" has launched new bath salts and towels and is hosting multiple campaigns and collaborative events. Its latest bath salts contain moisturizing ingredients such as placenta and hyaluronic acid, which are said to promote beautiful skin. The company also launched a skincare towel and began selling it online. It is also currently running an Instagram campaign.

[0034] The sentences that accept input can include the brand name (in the above example, Room A), information about the type, properties, and ingredients of products sold under that brand name (skin care bath additives, towels, beauty-related products, placenta, hyaluronic acid, honey, serum), and information about sales methods (online, e-commerce site, Instagram campaign), but are not limited to these and may also include other information. They may also be part of the above information.

[0035] The product information may be written in any format and does not need to be limited to a standard format. Such text is often prepared as sales promotion information when selling regular products or services, and for brand names used, explanatory information about the brand is often prepared in advance. In this embodiment, this already prepared text can be reused as product information to be entered, eliminating the need to take the trouble of preparing input information to use the system and reducing the burden on users.

[0036] Furthermore, the information that can be input is not limited to information (text information) related to a product that is directly input to the client terminal 102. For example, it is possible to accept input of link information (for example, a URL of image information such as a specific web page, video, or still image) from the client terminal 102. The server 101 can use this link information to obtain corresponding information from the link destination and accept information related to the product extracted from that information as input. When a web page is obtained from the link destination, the text information in the web page can be accepted as text input as is. A still image in the web page or another still image obtained from a link destination can be accepted as still image data as is.

[0037] Furthermore, for video images within a web page or video images obtained from other linked sites, the server 101 can capture still images from the video and accept them as input still image data. Still image data is included because this image information can also be vectorized and therefore can be used to determine similarity. The server 101 can also extract keywords, tag information, etc. from information transmitted from the client terminal 102 or information extracted from page information, and accept information obtained through internet searches or searches on social networking sites as input. Furthermore, the server 101 can extract text information contained in images within a web page or images obtained from other linked sites using OCR technology and accept it as input, or it can request an explanation of these images from the LLM and accept the resulting explanatory text information as input.

[0038] FIG. 5A shows an example of an input screen on the client terminal 102 side. Screen 500 displays an input area 501, and a client, who is a user of the client terminal 102, can input text such as that described above into the input area. Link information that the client requests to reference can also be input into input area 502. For example, link information for a web page, still image, or video related to the product input into input area 501 can be input. Still images and videos may also be directly input into input area 502 by a drag-and-drop operation. After inputting text, link information, an image, or the like, and clicking the recommended influencer search button 503, an influencer search request is transmitted to the server 101 together with the text information input into input area 501 and the information input into input area 502. In S401, the input of information transmitted from the client terminal 102 in this manner is accepted.

[0039] In the next step S402, the server 101 generates product information by integrating at least one of the pieces of information that can be accepted as input described in S401. The product information is summary information about the product generated by integrating information acquired by the server 101, and may include at least one of text information (sentence information) and image information. The server 101 generates the product information based on the text information, still images, and moving images transmitted from the client terminal 102, as well as page information acquired from linked sites based on link information, text information extracted from the page information, information acquired through keyword-based internet searches and SNS searches, character information extracted using OCR technology from images in web pages and other images acquired from linked sites, and descriptive text information about images acquired from LLM. The product information can be generated in the same manner as the summary information generated in S302. Specifically, the product information can be generated using mechanisms such as large language models (LLMs), recurrent neural networks (RNNs), or machine learning. For LLM, for example, a Transformer-based model (implemented in GPT-4) can be used.

[0040] The process of generating product information using LLM in S402 may be performed when it is necessary to integrate multiple types of information to generate summary information. For example, when only text information is received from the client terminal 102, the text information can be used as product information as is, and the generation process using LLM or the like may be omitted. Alternatively, when accepting information input in S401, the input content may be restricted in advance so that only input in a form that allows the generation process using LLM or the like to be omitted. In this case, for example, only input of text information into the input field 501 of FIG. 5A may be permitted.

[0041] In the next step S403, the server 101 vectorizes the product information generated in step S402. The vectorization is performed using a method similar to that used to vectorize the summary information in step S303 of Fig. 3, so that the vectorized product information has the same number of dimensions. As with the above, vectorization can be performed using, for example, OpenAI Embeddings (text-embedding-3), but is not limited to this. In any case, the number of dimensions is set to be the same as the result of vector processing of the summary information in step S303.

[0042] In the next step S404, the qualitative information in the qualitative information database 106 is compared with the vectorized product information generated in S403, and in the subsequent step S405, a first influencer is selected from the qualitative information with the highest similarity. The similarity can be determined based on the cosine similarity of the vector information. Cosine similarity is a measure of the similarity between two vectors being compared, and is expressed as the cosine value of the angle between the two vectors in vector space. Specifically, it can be calculated by dividing the dot product of two vectors (= the multiplication of vectors with both direction and magnitude) by the magnitude of the two vectors (= the L2 norm). This calculation normalizes the value to a range of -1 to 1, so a cosine similarity of 1 can be determined as "0 degrees, vectors facing the same direction = completely similar"; "0 degrees, vectors facing 90 degrees, independent / orthogonal directions = no relation to whether they are similar or not"; and "180 degrees, vectors facing opposite directions = completely dissimilar."

[0043] In the above, we have described the case where cosine similarity is used as a method for determining the similarity (= distance) of vectors, but the method for determining the similarity of vectors is not limited to that based on cosine similarity. For example, instead of cosine similarity, at least one of the following methods can be used.

[0044] Euclidean Distance: Euclidean distance measures the "straight-line" distance between two points. It is also used to calculate the distance between two points in multidimensional space, and depends on both the magnitude and direction of the vector. Manhattan Distance: Manhattan distance (or L1 distance) is calculated as the sum of the absolute values ​​of the differences in each dimension. It can be likened to the distance traveled along a grid of roads. Mahalanobis Distance: Mahalanobis distance is a distance measurement method that takes into account the distribution of data. It takes into account the correlation and variance of data and measures the "true" distance of multidimensional data. Jaccard Similarity: Jaccard Similarity measures the similarity between two sets. It considers two vectors as sets and evaluates the similarity by the ratio of their intersection and integration. Pearson Correlation Coefficient: The Pearson correlation coefficient measures the strength and direction of the linear relationship between two vectors. Its value ranges from -1 (perfect negative linear relationship) to 1 (perfect positive linear relationship). Spearman's Rank Correlation: Spearman's Rank Correlation measures the strength of a nonlinear relationship based on the ranks of two vectors. Hamming Distance: Hamming distance calculates the number of position-wise differences between two strings of the same length. It is mainly used for strings and binary data.

[0045] In S404, the qualitative information to be compared with the vectorized product information may be, for example, all of the qualitative information stored in the qualitative information database 106. Alternatively, the qualitative information to be compared may be narrowed down in advance before calculating the cosine similarity. The narrowing down may be performed, for example, by receiving a request for narrowing down conditions from the client terminal 102 and performing the narrowing down according to the requested narrowing down conditions. The narrowing down conditions may include, for example, a minimum number of followers, a specialty field (e.g., beauty or outdoors), or a target demographic (e.g., age group such as 18-34 years old or gender). Note that the narrowing down conditions are not limited to those listed here and may be freely input.

[0046] The narrowing down conditions may be specified in area 504 of screen 500 in Fig. 5B, which shows a modified example of screen 500 in Fig. 5A, before operating search button 503. When search button 503 is operated, the specified conditions are transmitted to server 101 together with the information of the text entered in input area 501 and the link information entered in input information 502.

[0047] When such narrowing down conditions are accepted, attribute information corresponding to the narrowing down conditions may be assigned to the qualitative information in advance and managed in the qualitative information database 106, and only the qualitative information whose attribute information matches the narrowing down conditions may be read out to calculate the cosine similarity.

[0048] In S405, the qualitative information having the highest cosine similarity calculated in S403 and having a cosine similarity equal to or greater than a predetermined threshold is identified, and the corresponding SNS user is designated as the first influencer. There may be one or more first influencers. If there are multiple first influencers, a predetermined number of SNS users whose cosine similarities are equal to or greater than the predetermined threshold are identified as first influencers in descending order. Cosine similarities are calculated, for example, as 0.82561928714675, 0.8216945111462, etc. Based on these cosine similarities, qualitative information having a similarity equal to or greater than a predetermined threshold (e.g., 0.8) is selected. The number of SNS users selected as first influencers may be a predetermined number, or all SNS users whose cosine similarities are equal to or greater than a predetermined value may be selected as first influencers. Alternatively, the top N% of SNS users may be selected as first influencers.

[0049] In the next step S406, the server 101 generates an explanatory document explaining why the selected SNS user is suitable for advertising the product or service corresponding to the product information, based on the summary information stored in the qualitative information database 106 for the SNS user selected in step S404 and the product information generated in step S402. The explanatory document can be generated, for example, by using a Transformer-based LLM to input the qualitative information and the product information together with a prompt for an explanation of the reasons for similarity.

[0050] In the following S407, the server 101 generates a result output screen based on the list of first influencers selected in S405 and the explanation document for the selection reasons generated in S406, and transmits the screen to the client terminal 102.

[0051] FIG. 5C shows an example of a result output screen on the client terminal 102 side. Screen 510 displays information 511 about the selected influencer. Information 511 includes the influencer's name, the influencer's cosine similarity, and a control 512 for displaying detailed information. On the result output screen, the selected influencers are displayed on the screen in descending order of cosine similarity. The client can confirm detailed information about the selected influencer by operating control 512 of information 511. Note that although the cosine similarity is displayed with a large number of decimal points, the number of digits may be narrowed down. The information 511 may also include a list of multiple influencers divided by percentiles of similarity. For example, the selected influencers may be grouped by the magnitude of the cosine similarity and displayed for each group. The groups may be divided into three groups, for example, those with high, medium, and low similarity.

[0052] 5D shows an example of the display of detailed information about "Beauty Professional X" displayed at the top. On screen 520 in FIG. 5C, when control 512 is clicked, detailed reasons for selecting Beauty Professional X are displayed in display area 521. Here, the reasons for selection are displayed in relation to items such as matching with the target demographic of the advertisement, the influencer's expertise and influence, similarity with the influencer's concept, characteristics of the influencer's content, and collaboration with the influencer's campaign, for example.

[0053] A scroll bar 522 is displayed on screen 520, and if you want to check detailed information about other selected influencers, you can operate scroll bar 522 to scroll the screen to the position where detailed information about the other selected influencers is displayed. After obtaining a list of selected influencers in the manner described above, additional conditions may be added to further refine the list. In this case, additional conditions may include pre-defined conditions as well as arbitrarily entered conditions, such as "Please introduce influencers who are more suitable for younger generations." An example of the display in this case is shown in FIG. 5E. In FIG. 5E, an input area 530 displaying the refinement conditions 504 and search button 503 of FIG. 5B is superimposed on the screen 510 of FIG. 5C. When the refinement conditions are entered in the input area 530 and the search button 503 is operated, the refinement conditions are sent to the server 101, and the refinement process is executed.

[0054] In the present embodiment described above, the information on influencers to be compared with product information is vectorized in advance and prepared as qualitative information, thereby reducing the processing load when selecting influencers.

[0055] Furthermore, according to this embodiment, the relevance of an influencer is determined based on the cosine similarity between vectorized product information and vectorized influencer information (qualitative information). In this embodiment, product-related information is vectorized and compared with qualitative information, so the form of input information about the product is not specified, and any information (text, image, video, etc.) can be used as input. Therefore, unlike in the past, product information does not need to be quantified and keyworded in the same format to enable comparison and search by quantifying and keywording influencer characteristics, which reduces the user's workload.

[0056] Furthermore, this embodiment also makes it possible to mechanically select candidates using existing information, such as text data on product introduction web pages. Furthermore, by improving the accuracy of product information, it becomes possible to select more relevant influencers. Clients can easily obtain influencer candidates suitable for advertising and written reasons for their selection simply by entering information about products or services they are familiar with, enabling them to instantly obtain results of similar quality to those obtained by relying on experts. Furthermore, by setting filtering conditions, it becomes possible to further improve the accuracy of influencer selection.

[0057] The following is another possible extension of this embodiment. For example, LLM can be used to input text information along with a prompt requesting revisions to the text information, allowing the user to make revisions. For example, a prompt such as, "Below is the text information about the product you would like an influencer to introduce, along with a request for the characteristics of the influencer who will introduce the product. Based on this, please create a product introduction that clearly defines the characteristics you are looking for in an influencer." This allows revisions to be made in accordance with the instructions. Using the revised text information as input information in S401 of Figure 4 and performing the processing of this embodiment, it becomes possible to select an influencer that better meets the client's needs.

[0058] <Summary of the embodiment> The above-described embodiment discloses at least the following information processing device and computer program. (1) A server comprising one or more processors, a memory, and a program stored in the memory, the program, when executed by the one or more processors, causing the server to: obtaining vectorized influencer information for each of a plurality of influencers; Obtaining product information relating to a product that is a product or service being advertised; Vectorizing the product information; Calculating a similarity between the vectorized influencer information and the vectorized product information; selecting a first influencer having a high degree of relevance to the product from among the plurality of influencers based on the calculated similarity; outputting information about the first influencer; The server that runs the (2) The program, when executed by the one or more processors, causes the server to: collecting information about the plurality of influencers; generating influencer information about each of the plurality of influencers from the collected information about each of the plurality of influencers; vectorizing the influencer information; storing the vectorized influencer information; Execute The server according to (1), wherein obtaining vectorized influencer information for each of the plurality of influencers is performed by obtaining the stored influencer information. (3) The server according to (2), wherein acquiring the stored influencer information includes acquiring the influencer information of influencers that satisfy specified conditions. (4) A server according to any one of (1) to (3), wherein collecting information about the plurality of influencers includes at least one of obtaining information about each influencer from an external server that provides a social network service via a network, and obtaining response information to a predetermined questionnaire from each influencer. (5) A server described in any one of (1) to (4), wherein the plurality of influencers are registered users of a specified social networking service who disseminate information on the specified social networking service and whose number of followers, posts, views, and likes exceeds a specified number. (6) The server according to any one of (1) to (5), wherein the influencer information includes at least one of profile information, information about the content of the post, and information about the gender and age distribution of viewers of the post. (7) The server according to any one of (1) to (6), wherein the vectorized influencer information and the vectorized product information have the same number of dimensions. (8) The server according to any one of (1) to (7), wherein the selection of the first influencer is performed by selecting an influencer from among the plurality of influencers whose calculated similarity is greater than a predetermined threshold. (9) The server according to any one of (1) to (8), wherein the similarity is calculated based on at least one of cosine similarity, Euclidean distance, Manhattan distance, Mahalanobis distance, Jaccard similarity, Pearson correlation coefficient, Spearman rank correlation coefficient, and Hamming distance. (10) The server according to any one of (1) to (9), wherein the product information includes at least one of text information in any format and image information relating to the product. (11) Obtaining the product information about the product includes: receiving information about the product, including at least one of first text information, first image information, and link information, from a client terminal connected to the server via a network; When the link information is included in the information about the product, acquiring at least one of page information and second image information based on the link information; When the page information is acquired, performing a search based on a keyword related to the product extracted from the page information to acquire search information; obtaining second text information extracted from at least one of the first image information and the second image information; generating the product information based on at least one of the first text information, the first image information, the page information, the second image information, the search information, and the second text information; Including, The server according to any one of (1) to (10), wherein outputting the information of the first influencer includes transmitting the information of the first influencer to the client terminal. (12) The program, when executed by the one or more processors, causes the server to: The server according to any one of (1) to (11), further configured to generate explanatory information on the reason for selecting the selected first influencer. (13) The server according to (12), wherein the explanatory information is generated using a Transformer-based large-scale language model based on the influencer information that is not vectorized and the product information that is not vectorized. (14) A program for causing a computer to function as a server according to any one of (1) to (13).

[0059] [Other embodiments] The invention is not limited to the above-described embodiments, and various modifications and variations are possible within the spirit and scope of the invention. Therefore, the following claims are appended to clarify the scope of the invention. The information processing device according to the present invention can also be realized by a computer program that causes one or more computers to function as the information processing device. The computer program can be provided / distributed by being recorded on a computer-readable recording medium or via a telecommunications line. [Explanation of symbols]

[0060] 10: System, 101: Server, 102: Client terminal, 103: SNS server, 104: Network

Claims

1. A server comprising one or more processors, a memory, and a program stored in the memory, the program, when executed by the one or more processors, causing the server to: obtaining vectorized influencer information for each of a plurality of influencers; Acquiring product information relating to a product or service that is an advertised product; and vectorizing the product information. Calculating a similarity between the vectorized influencer information and the vectorized product information; selecting a first influencer having a high degree of relevance to the product from among the plurality of influencers based on the calculated similarity; outputting information of the first influencer; The server that runs the

2. The program, when executed by the one or more processors, causes the server to: collecting information about the plurality of influencers; generating influencer information about each of the plurality of influencers from the collected information about each of the plurality of influencers; vectorizing the influencer information; storing the vectorized influencer information; Execute The server of claim 1 , wherein obtaining vectorized influencer information for each of the plurality of influencers is performed by obtaining the stored influencer information.

3. The server of claim 2 , wherein the acquiring the stored influencer information includes acquiring the influencer information of influencers that satisfy specified conditions.

4. The server of claim 3, wherein collecting information about the plurality of influencers includes at least one of obtaining information about each influencer from an external server that provides a social network service via a network, and obtaining response information to a predetermined questionnaire from each influencer.

5. The server of claim 4, wherein the plurality of influencers are registered users of a specified social networking service who disseminate information on the specified social networking service and whose number of followers, posts, views, and likes exceeds a specified number.

6. The server according to claim 5 , wherein the influencer information includes at least one of profile information, information about the content of posts, and information about the gender and age ratio of viewers of posts.

7. The server according to claim 6 , wherein the vectorized influencer information and the vectorized product information have the same number of dimensions.

8. The server according to claim 7 , wherein the first influencer is selected by selecting, from the plurality of influencers, an influencer whose calculated similarity is greater than a predetermined threshold.

9. 9. The server according to claim 8, wherein the similarity is calculated based on at least one of cosine similarity, Euclidean distance, Manhattan distance, Mahalanobis distance, Jaccard similarity, Pearson correlation coefficient, Spearman's rank correlation coefficient, and Hamming distance.

10. The server according to claim 9 , wherein the product information includes at least one of text information in any format and image information relating to the product.

11. The obtaining of the product information relating to the product includes: receiving information about the product, including at least one of first text information, first image information, and link information, from a client terminal connected to the server via a network; When the link information is included in the information about the product, acquiring at least one of page information and second image information based on the link information; When the page information is acquired, performing a search based on a keyword related to the product extracted from the page information to acquire search information; obtaining second text information extracted from at least one of the first image information and the second image information; generating the product information based on at least one of the first text information, the first image information, the page information, the second image information, the search information, and the second text information; Including, The server of claim 10 , wherein outputting the information of the first influencer includes transmitting the information of the first influencer to the client terminal.

12. The program, when executed by the one or more processors, causes the server to: The server of claim 11 , further configured to generate explanatory information about the reason for the selection of the selected first influencer.

13. The server according to claim 12 , wherein the description information is generated using a Transformer-based large-scale language model based on the influencer information that is not vectorized and the product information that is not vectorized.

14. A program for causing a computer to function as the server according to any one of claims 1 to 13.

Images