Information processing system, information processing method, and information processing program

Abstract

The present invention improves precision of filtering of content. An information processing system 1 includes a first extraction unit 102 that extracts first BCT component information related to a component of a first behavior change technique (BCT) included in first content determined to be an object of filtering, a second extraction unit 103 that extracts second BCT component information related to a component of a second behavior change technique (BCT) included in second content that is not determined to be an object of filtering, and a generating unit 104 that estimates reference information serving as a determination reference for determining whether or not the content is content that is an object of filtering, on the basis of the first BCT component information and the second BCT component information, and generates a model based on the reference information.

Description

Information processing system, information processing method, and information processing program CROSS-REFERENCE TO RELATED APPLICATIONS

[0001] This application is based on Japanese Patent Application No. 2023-132015 filed on August 14, 2023, the contents of which are incorporated herein by reference.

[0002] The present disclosure relates to an information processing system, an information processing method, and an information processing program.

[0003] It has been known in the past to filter various types of information using models based on machine learning. For example, Patent Literature 1 describes a method in which learning data for a Bayesian filter is managed as a learning spam filter, and spam messages are identified using the learning spam filter.

[0004] Patent No. 5676709

[0005] However, for example, in the conventional spam message filter described in Patent Document 1, there is a risk that emails that are not spam messages may be mistakenly determined to be spam messages (false positives), or that spam messages may mistakenly slip through the filter (false negatives). Such problems may occur not only in spam message filters, but also in filters for various content. Therefore, one object of the present disclosure is to provide an information processing system, an information processing method, and an information processing program that can improve the accuracy of content filtering.

[0006] An information processing system according to one aspect of the present disclosure includes a first extraction unit that extracts first BCT component information relating to components of a first behavioral change technique (first BCT) contained in first content determined to be subject to filtering, a second extraction unit that extracts second BCT component information relating to components of a second behavioral change technique (second BCT) contained in second content that is not determined to be subject to filtering, and a generation unit that estimates reference information that serves as a criterion for determining whether or not content is subject to filtering based on the first BCT component information and the second BCT component information, and generates a model based on the reference information.

[0007] An information processing method according to another aspect of the present disclosure includes the steps of: extracting, by an information processing device, first BCT component information relating to components of a first behavioral change technique (first BCT) contained in first content determined to be subject to filtering; extracting second BCT component information relating to components of a second behavioral change technique (second BCT) contained in second content not determined to be subject to filtering; and estimating, based on the first BCT component information and the second BCT component information, reference information that serves as a criterion for determining whether or not the content is subject to filtering, and generating a model based on the reference information.

[0008] An information processing program according to another aspect of the present disclosure causes a computer to execute the steps of: extracting first BCT component information relating to components of a first behavioral change technique (first BCT) contained in first content determined to be subject to filtering; extracting second BCT component information relating to components of a second behavioral change technique (second BCT) contained in second content not determined to be subject to filtering; and estimating, based on the first BCT component information and the second BCT component information, reference information that serves as a criterion for determining whether or not the content is subject to filtering, and generating a model based on the reference information.

[0009] According to the present disclosure, the accuracy of content filtering can be improved.

[0010] FIG. 1 is a diagram showing an example of BCT classification according to this embodiment. FIG. 2A is a conceptual diagram of generation of a filtering model according to the prior art. FIG. 2B is a conceptual diagram of generation of a filtering model according to this embodiment. FIG. 3 is a diagram showing an example of the hardware configuration of each information processing device in an information processing system according to this embodiment. FIG. 4 is a diagram showing an example of the functional configuration of the information processing system according to this embodiment. FIG. 5A is a diagram showing an example of first BCT component information according to this embodiment. FIG. 5B is a diagram showing an example of second BCT component information according to this embodiment. FIG. 6 is a flowchart showing an example of the operation of a generation device according to this embodiment. FIG. 7 is a flowchart showing an example of the operation of a filter device according to this embodiment.

[0011] Embodiments of the present disclosure will be described with reference to the accompanying drawings, in which the same reference numerals denote the same or similar configurations.

[0012] (Overview) In an information processing system according to this embodiment, an information processing device extracts information (hereinafter referred to as "first BCT component information") regarding components of a first behavior modification technique (first BCT) contained in first content determined to be subject to filtering. The information processing device extracts information (hereinafter referred to as "second BCT component information") regarding components of a second behavior modification technique (first BCT) contained in second content not determined to be subject to filtering. Based on the first BCT component information and the second BCT component information, the information processing device estimates information (hereinafter referred to as "reference information") that serves as a criterion for determining whether or not content is subject to filtering, and generates a model based on the reference information. Note that extraction can also be referred to as encoding, etc.

[0013] In this embodiment, a behavior change technique (BCT) is a technique or method that affects (influences) behavior change. For example, BCTTv1 (Michie S, Richardson M, Johnston M, et al.: The behavior change technique taxonomy (v1) of 93 hierarchically clustered techniques: building an international consensus for the reporting of behavior change interventions. Ann Behav Med 2013; 46: 81-95.) defines 93 BCTs in 16 groups. Note that the definition of BCTs is not limited to BCTTv1 and may be defined in any way as long as it comprehensively covers behavior change techniques. Furthermore, the BCT according to this embodiment includes a group containing one or more BCTs (hereinafter referred to as a "BCT group"), and the terms "BCT group" and "BCT" may be interchangeable.

[0014] FIG. 1 is a diagram showing an example of BCT classification according to this embodiment. One or more BCTs are classified according to a predetermined taxonomy. As shown in FIG. 1, BCTTv1 defines 16 BCT groups: "1. Goals and planning," "2. Feedback and monitoring," "3. Social support," "4. Shaping knowledge," "5. Natural consequences," "6. Comparison of behavior," "7. Associations," "8. Repetition and substitution," "9. Comparison of outcomes," "10. Reward and threat," "11. Regulation," "12. Antecedents," "13. Identity," "14. Scheduled consequences," "15. Self-belief," and "16. Covert learning."

[0015] In BCTTv1, one or more BCTs belong to each of the 16 BCT groups shown in FIG. 1. For example, BCTs such as "5.5. Anticipated regret" belong to the BCT group "5. Natural consequences." Also, BCTs such as "10.11. Future punishment" belong to the BCT group "10. Reward and threat." Although not shown, one or more BCTs also belong to the other groups.

[0016] Each BCT has a component, and the degree of each BCT contained in a certain content may be indicated as a component value. Also, the sum of the component values ​​of each BCT belonging to the same BCT group may be indicated as the component value of the BCT group.

[0017] 2A and 2B are conceptual diagrams illustrating the generation of a filtering model according to the prior art and the present embodiment. In FIG. 2B, the content to be filtered is assumed to be spam and / or junk messages, the first content determined to be subject to filtering is a spam message, and the second content not determined to be subject to filtering is a message that is not a spam message (hereinafter referred to as a "normal message"). However, FIG. 2B is merely an example, and the content to be filtered and the first content are not limited to spam messages, and the second content is not limited to normal messages.

[0018] Here, a nuisance message is a message intentionally designed to hinder a user's sound decision-making by exaggerating the costs of not acting quickly or by placing psychological burdens on the user with language that incites feelings of inferiority, without providing the user (e.g., the message reader) with information that will help them make better decisions. The nuisance message may be information transmitted through various media, such as e-mail, short messages, or stealth marketing sites known as "How did you like it?" sites. The nuisance message may also be at least one of text data, image data, and audio data.

[0019] Furthermore, spam messages can be considered information that contains excessive amounts of a specific BCT. The specific BCT may be, for example, a BCT belonging to at least one of the following categories: sludge, threats, and incitement (e.g., "5.5. Anticipated regret" and / or "10.11. Future punishment" in Figure 1). Sludge refers to the selfish actions of companies, fraud groups, etc., which lead users to behave in a way that is disadvantageous to them or prevent users from behaving in a way that is advantageous to them. For example, in the context of spam messages, sludge may be encouraging users to enter their credit card numbers into a fraudulent website or not giving users enough time to consult with their family before transferring money to cover a fictitious bill.

[0020] For example, as shown in Figure 2A, in one example of conventional technology, a judgment criterion based on Bayesian estimation is generated by machine learning using training data annotated as spam messages, and a filtering model based on the judgment criterion is generated. The judgment criterion based on Bayesian estimation estimates the probability of a message being spam using conditional probability, and determines that a message is spam if it exceeds a preset threshold. Note that "annotation" may also be referred to as tagging, labeling, or the like.

[0021] Other known conventional filtering technologies based on training data annotated as spam messages include neural networks and random forests. Blocking technologies based on pre-set lists (e.g., blacklists, real-time blacklists, reputation lists, etc.) are also known. However, as spam messages become more personalized and sophisticated using generative AI, models using the above filtering and blocking technologies may increase the probability of false positives and false negatives, reducing the accuracy of filtering.

[0022] 2B , in the information processing system according to this embodiment, first BCT component information relating to the first BCT components contained in the first teacher data annotated as a nuisance message is extracted, and second BCT component information relating to the second BCT components contained in the second teacher data annotated as a normal message is extracted. By analyzing the difference between the first BCT component information and the second BCT component information, reference information that serves as a criterion for determining whether a message is a nuisance message is estimated, and a model for filtering nuisance messages based on the reference information is generated.

[0023] In this manner, in the information processing system according to this embodiment, first BCT component information is extracted from first content determined to be subject to filtering (e.g., first teacher data annotated as a spam message), and second BCT component information is extracted from second content determined not to be subject to filtering (e.g., second teacher data annotated as a normal message). Furthermore, based on a comparison of the extracted first BCT component information and second BCT component information, reference information is estimated as a criterion for determining whether or not content is subject to filtering (e.g., a spam message), and a model based on this reference information is generated. Therefore, compared with conventional criteria, a model based on a criterion with a lower probability of false positives and false negatives can be generated, thereby improving the accuracy of filtering.

[0024] (Configuration) Next, the configuration of the information processing system 1 according to this embodiment will be described. The information processing system 1 may be configured to include a single information processing device, or may be configured to include multiple information processing devices. For example, the information processing system 1 may be configured to include multiple information processing devices that implement some of the functional configurations described below, and the functional configurations described below may be realized by the multiple devices connected via a wired and / or wireless network.

[0025] 3 is a diagram showing an example of the hardware configuration of each information processing device constituting the information processing system according to this embodiment. Each information processing device (e.g., a generating device 10 and a filter device 20 described below) in the information processing system 1 has a processor 11 such as a CPU (Central Processing Unit) corresponding to an arithmetic unit, a storage device 12, a communication device 13, and an input / output device 14. These components are connected via a bus so as to be able to transmit and receive data to and from each other.

[0026] The processor 11 is, for example, a CPU, and controls each information processing device in the information processing system 1. The processor 11 may execute various processes described in this embodiment by reading and executing a program from the storage device 12. Each information processing device in the information processing system 1 may include one or more processors 11. Furthermore, each of the devices may be referred to as a computer.

[0027] The storage device 12 is configured by, for example, storage such as a memory, a hard disk drive (HDD), and / or a solid state drive (SSD), etc. The storage device 12 may store various information necessary for the processor 11 to execute processing (for example, an information processing program executed by the processor 11, etc.).

[0028] The communication device 13 is a device that communicates via a network and may include, for example, a network card, a communication module, a chip, an antenna, etc. The communication device 13 may include a transmitting unit that transmits various information via the network and / or a receiving unit that receives various information via the network.

[0029] The input / output device 14 includes an input unit such as a keyboard, a touch panel, a mouse, and / or a microphone, and an output unit such as a display and / or a speaker. The input unit accepts various types of information input from a user. The output unit may also include a display unit that displays various types of information to the user.

[0030] The hardware configuration described above is merely an example. Each information processing device in the information processing system 1 may omit some of the hardware shown in FIG. 4 or may include hardware not shown in FIG. 4. The hardware shown in FIG. 4 may be configured with one or more chips. When the information processing device is configured with multiple devices, each device may include at least some of the hardware.

[0031] Fig. 4 is a diagram showing an example of the functional configuration of an information processing system according to this embodiment. As shown in Fig. 4, the information processing system 1 includes an information processing device (hereinafter referred to as "generation device") 10 that generates the model (e.g., Fig. 2B), and an information processing device (hereinafter referred to as "filter device") 20 that acquires content and filters the acquired content based on the model. Note that the generation device 10 and the filter device 20 may be configured as the same information processing device or as separate information processing devices.

[0032] 4 is merely an example, and it goes without saying that the generating device 10 and the filtering device 20 may each have functional configurations not shown. The functions of each functional configuration are realized by loading an information processing program into hardware such as the processor 11 shown in FIG. 3, and operating at least one of the storage device 12, the communication device 13, and the input / output device 14 under the control of the processor 11.

[0033] As shown in FIG. 4 , the generating device 10 includes a storage unit 101 , a first extraction unit 102 , a second extraction unit 103 , and a generating unit 104 .

[0034] The storage unit 101 stores first content determined to be subject to filtering and second content not determined to be subject to filtering. Specifically, the storage unit 101 may store first training data annotated as subject to filtering (e.g., training data annotated as spam messages) as the first content. The storage unit 101 may also store second training data annotated as not subject to filtering (e.g., training data annotated as normal messages) as the second content.

[0035] The storage unit 101 may also store a model generated by the generation unit 104. The storage unit 101 may also store a model distilled based on the model generated by the generation unit 104.

[0036] The first extraction unit 102 extracts first BCT component information related to components of the first BCT included in the first content determined to be a filtering target. Specifically, the first extraction unit 102 may extract the first BCT component information by machine learning using the first content (e.g., first training data annotated as a filtering target).

[0037] The first BCT component information may be a component table indicating fixed values ​​based on an average of the first BCT components included in the first content, or may indicate a distribution of the first BCT components included in the first content. The first BCT component information may also include at least one of information regarding the ratio of the information amount of the first BCT components to the information amount of the first content (hereinafter referred to as "first content ratio information") and information regarding the ratio of the information amount of a specific BCT component to the information amount corresponding to the first BCT component (hereinafter referred to as "first composition ratio information").

[0038] 5A is a diagram showing an example of first BCT component information according to this embodiment. For example, FIG. 5A shows an example of first BCT component information extracted from first teacher data annotated as a spam message. Also, in FIG. 5A, the first BCT component information is a component table showing fixed values ​​based on the average of the components of the first teacher data. However, as described above, the first BCT component information may also show the distribution of the components of the first teacher data.

[0039] For example, in FIG. 5A , the first BCT component information indicates that, assuming the information amount of the first teacher data annotated as spam is 100%, the information amount of the threatening BCT components is 50%, the information amount of the provocative BCT components is 40%, the information amount of the other BCT components is 5%, and the information amount of the components not belonging to any BCT (no BCT) is 5%. In this case, the information amount of the first BCT components (here, the threatening BCT, the provocative BCT, and the other BCTs) included in the first teacher data is 95%, so the first content ratio information may indicate 95 / 100. Furthermore, since the information amount of the first BCT components is 95%, and the information amount of the specific BCT components (here, the threatening BCT and the provocative BCT) is 90% (= 40% + 50%), the first composition ratio information may indicate 90 / 95.

[0040] The second extraction unit 103 extracts second BCT component information related to components of the second BCT included in the second content that has not been determined to be subject to filtering. Specifically, the second extraction unit 103 may extract the second BCT component information by machine learning using the second content (e.g., second training data annotated as not subject to filtering).

[0041] The second BCT component information may be a component table showing fixed values ​​based on an average of the second BCT components included in the second content, or may show a distribution of the second BCT components included in the second content. Furthermore, the second BCT component information may include at least one of information on the ratio of the information amount of the second BCT components to the information amount of the second content (hereinafter referred to as "second content rate information") and information on the ratio of the information amount of a specific BCT component to the information amount corresponding to the second BCT component (hereinafter referred to as "second composition ratio information").

[0042] 5B is a diagram showing an example of second BCT component information according to this embodiment. For example, FIG. 5B shows an example of second BCT component information extracted from second teacher data annotated as a normal message. Also, in FIG. 5B, the second BCT component information is a component table showing fixed values ​​based on the average of the components of the second teacher data. However, as described above, the second BCT component information may also show the distribution of the components of the second teacher data.

[0043] For example, in FIG. 5B , the second BCT component information indicates that, assuming the amount of information in the second teacher data annotated as normal email is 100%, the amount of information not belonging to any BCT (no BCT) is 70%, the amount of information in the comparison BCT is 10%, the amount of information in the planning BCT is 10%, and the amount of information in the other BCTs is 10%. In this case, the amount of information corresponding to the components of the second BCT (here, the comparison BCT, the planning BCT, and the other BCTs) included in the second teacher data is 30%, so the second content ratio information may indicate 30 / 100. Furthermore, the amount of information corresponding to the components of the second BCT is 30%, and the amount of information corresponding to the components of specific BCTs (here, the threatening BCT and the provocative BCT) is 0%, so the second composition ratio information may indicate 0.

[0044] The generation unit 104 estimates reference information that serves as a criterion for determining whether content is subject to filtering, based on the first BCT component information and the second BCT component information, and generates a model based on the reference information. The reference information may include at least one of information on the ratio of the amount of information of a component of a third behavior modification technique (third BCT) contained in the content to the amount of information of the content subject to filtering (hereinafter referred to as "content ratio information"), information on the ratio of the amount of information of a component of a specific BCT to the amount of information of the component of the third BCT (hereinafter referred to as "composition ratio information"), and information on the specific BCT (hereinafter referred to as "BCT information").

[0045] Specifically, the generation unit 104 may estimate content rate information regarding the ratio of the information amount of the third BCT components contained in the content to the information amount of the content to be filtered, based on the first content rate information and the second content rate information. For example, as described in Figures 5A and 5B, if the first content rate information is 95 / 100 and the second content rate information is 30 / 100, the generation unit 104 may estimate the content rate information, which is the criterion for determining whether the content is to be filtered (i.e., a spam message), to be 50 / 100.

[0046] Furthermore, the generation unit 104 may estimate composition ratio information regarding the ratio of the information amount of a specific BCT component to the information amount of a third BCT component, based on the first composition ratio information and the second composition ratio information. For example, as described in Figures 5A and 5B, if the first composition ratio information is 90 / 95 and the second composition ratio information is 0, the generation unit 104 may estimate the composition ratio information, which is the criterion for determining whether content is subject to filtering (i.e., a spam message), to be 40 / 50.

[0047] Furthermore, the generation unit 104 may estimate BCT information indicating a specific BCT based on the first BCT component information and the second BCT component information. For example, as shown in Figures 5A and 5B, the first BCT component information contains a higher proportion of threatening BCT and provocative BCT components than the proportion of comparison BCT and planning BCT components, which are often contained in normal messages. The generation unit 104 may estimate BCT information indicating a specific BCT that tends to be contained in content to be filtered (i.e., spam messages) based on a comparison between the first BCT component information and the second BCT component information.

[0048] As shown in FIG. 4 , the filter device 20 includes an acquisition unit 201 , a filter unit 202 , and an output unit 203 .

[0049] The acquisition unit 201 acquires content. Specifically, the acquisition unit 201 acquires content to be determined as to whether it is subject to filtering (for example, a target message to be determined as to whether it is a spam message).

[0050] The filter unit 202 determines whether the content acquired by the acquisition unit 201 is content to be filtered based on the model generated by the generation device 10, and performs filtering based on the determination result. Note that the filter unit 202 may use the model generated by the generation unit 104 itself, or may use a model obtained by distilling the model. As described above, the model may use at least one of the content rate information and the composition ratio information as a determination criterion.

[0051] Specifically, the filter unit 202 may determine whether or not the content is a filtering target by using a model based on the content content content amount ratio and the content content content ratio information relative to the content content amount acquired by the acquisition unit 201. For example, the model may determine whether or not the content is a filtering target by whether or not the content content content amount ratio relative to the content content amount acquired by the acquisition unit 201 exceeds a value indicated by the content content content ratio information (e.g., 50 / 100), or by how much the content content content deviates from the value indicated by the content content content ratio information.

[0052] Furthermore, the filter unit 202 may determine whether or not the content is a filtering target by using a model based on the composition ratio information and the ratio of the information amount of a specific BCT component to the information amount of BCT included in the content acquired by the acquisition unit 201. For example, the model may determine whether or not the content is a filtering target based on whether or not the ratio of the specific BCT component to the information amount of BCT included in the content acquired by the acquisition unit 201 exceeds a value indicated by the composition ratio information (e.g., 40 / 50), or on how much the ratio deviates from the value indicated by the composition ratio information.

[0053] In addition, the filter unit 202 may determine whether the content is subject to filtering by using a model based on the ratio of the amount of information of a specific BCT component to the amount of information of the BCT contained in the content acquired by the acquisition unit 201 and the above-mentioned content rate information and composition ratio information.

[0054] The output unit 203 outputs information regarding the results of filtering by the filter unit 202 (hereinafter referred to as "filtering information"). The filtering information may be fed back to the generating device 10. For example, if content filtered by the filter unit 202 as content to be filtered is content that should not be filtered (i.e., in the case of a false positive), the filtered content may be annotated as content that should not be filtered and fed back to the generating device 10 as second training data. On the other hand, if content not filtered by the filter unit 202 as content to be filtered is content that should be filtered (i.e., in the case of a false negative), the unfiltered content may be annotated as content to be filtered and fed back to the generating device 10 as second training data. This further improves the accuracy of filtering based on the model generated by the generating device 10. (Operation) Next, the operation of the information processing system 1 according to this embodiment will be described. The operations of the generating device 10 and the filter device 20 shown in FIGS. 6 and 7 are merely examples and are not limited to those shown. For example, some steps may be deleted and / or added, or some steps (for example, steps S101 and S102 in FIG. 6) may be executed in a different order or in parallel.

[0055] 6 is a flowchart showing an example of the operation of the generating device 10 according to the present embodiment. As shown in FIG. 6, the generating device 10 extracts first BCT component information related to components of the first BCT included in the first content determined to be subject to filtering (step S101). For example, as described with reference to FIGS. 2B and 5A, the generating device 10 may extract the first BCT component information by machine learning using first training data annotated as spam.

[0056] The generating device 10 extracts second BCT component information related to components of the second BCT included in the second content that has not been determined to be subject to filtering (step S102). For example, as described with reference to FIGS. 2B and 5B , the generating device 10 may extract the second BCT component information by machine learning using second training data annotated as normal email.

[0057] The generating device 10 estimates reference information that serves as a criterion for determining whether or not content is subject to filtering, based on the first BCT component information extracted in step S101 and the second BCT component information extracted in step S102 (step S103). Specifically, as described with reference to Figures 2B, 5A, and 5B, the generating device 10 may analyze the difference between the first BCT component information and the second BCT component information to estimate the reference information that serves as the criterion. The generating device 10 generates a model based on the reference information estimated in step S103 (step S104).

[0058] 7 is a flowchart showing an example of the operation of the filter device 20 according to this embodiment. As shown in FIG. 7, the filter device 20 acquires content to be determined as to whether it is a filtering target (step S201). For example, the filter device 20 acquires a target message to be determined as to whether it is a spam message.

[0059] The filter device 20 determines whether the content acquired in step S201 is a content to be filtered based on the model generated by the generation device 10, and performs filtering based on the determination result (step S202). For example, the filter device 20 determines whether the target message is a spam message based on the model, and performs filtering based on the determination result.

[0060] The filter device 20 outputs filtering information relating to the result of the filtering in step S202 (step S203).

[0061] As described above, in the information processing system 1 according to this embodiment, the generation device 10 estimates reference information as a criterion for determining whether or not content is subject to filtering (e.g., a spam message) based on the first BCT component information extracted from the first content determined to be subject to filtering and the second BCT component information extracted from the second content determined not to be subject to filtering, and generates a model based on the reference information. Therefore, it is possible to generate a filtering model based on a criterion with a lower probability of false positives and false negatives compared to conventional criteria.

[0062] Furthermore, in the information processing system 1 according to the present embodiment, it is determined whether or not the acquired content is subject to filtering based on the model generated by the generation device 10, and filtering is performed based on the determination result, thereby improving the accuracy of filtering.

[0063] (Other Embodiments) In the above embodiment, spam messages were used as an example of content to be filtered, but the content to be filtered is not limited to this. Any content can be filtered, as long as there is a combination of first content that is highly likely to contain a BCT that indicates a significant trend and second content that is less likely to contain the BCT that indicates a significant trend, such as a catchphrase, an advertisement, or content designed by prompt engineering. For example, the combination of the first content and the second content may be determined based on whether the catchphrase is well-written, whether the advertisement is well-balanced, or whether the content designed by prompt engineering is appropriate or inappropriate.

[0064] Furthermore, the content to be filtered may be "content that attempts to encourage a specific action." For example, the generating device 10 may generate a model that determines whether a coercive email is intended to encourage a person in a weaker position to take action, and the filtering device may use the model to determine whether the target content is intended to encourage a person in a weaker position to take action. Such a model may be used to determine whether or not there is sexual harassment, power harassment, or bullying of subcontractors.

[0065] Furthermore, the specific BCT may be any BCT that is expected to be more prominently included in content determined to be subject to filtering than in content determined not to be subject to filtering. For example, when content determined to be well-written is targeted for filtering, the specific BCT may be, for example, but is not limited to, a BCT that presents an objective perspective on the target behavior. When content determined to be well-balanced is targeted for filtering, the specific BCT may be, for example, but is not limited to, a BCT that presents a new perspective. When content designed by prompt engineering and determined to be appropriate is targeted for filtering, the specific BCT may be, for example, but is not limited to, a BCT that provides feedback.

[0066] In addition, the numbers assigned to the first BCT, second BCT, and third BCT in this embodiment are assigned to distinguish the three types of BCTs and do not indicate the order. Therefore, as long as each BCT is distinguishable, the numbers first, second, and third may be interchanged as appropriate.

[0067] The above-described embodiments are intended to facilitate understanding of the present disclosure and are not intended to limit the present disclosure. The flowcharts, sequences, elements included in the embodiments, as well as their arrangements, materials, conditions, shapes, sizes, and the like, are not limited to those exemplified and may be modified as appropriate.

Claims

1. An information processing system comprising: a first extraction unit that extracts first BCT component information regarding components of a first behavior change technique (first BCT) contained in first content determined to be subject to filtering; a second extraction unit that extracts second BCT component information regarding components of a second behavior change technique (second BCT) contained in second content not determined to be subject to filtering; and a generation unit that estimates reference information that serves as a criterion for determining whether or not content is subject to filtering based on the first BCT component information and the second BCT component information, and generates a model based on the reference information.

2. The information processing system of claim 1, wherein the reference information includes at least one of content rate information regarding the ratio of the amount of information of a component of a third behavior modification technique (third BCT) contained in the content to the amount of information of the content to be filtered, composition ratio information regarding the ratio of the amount of information of a component of a specific BCT to the amount of information of the components of the third BCT, and BCT information regarding the specific BCT.

3. An information processing system as described in claim 2, wherein the first BCT component information includes at least one of first content ratio information regarding the ratio of the amount of information of the first BCT component to the amount of information of the first content, and first composition ratio information regarding the ratio of the amount of information of the specific BCT component to the amount of information corresponding to the first BCT component, and the second BCT component information includes at least one of second content ratio information regarding the ratio of the amount of information of the second BCT component to the amount of information corresponding to the second BCT component.

4. An information processing system as described in any one of claims 1 to 3, further comprising: an acquisition unit that acquires content; and a filter unit that determines whether the acquired content is the content to be filtered based on the model, and performs filtering based on the determination result.

5. The information processing system according to claim 4, wherein the first content is a spam message, the second content is a normal message other than the spam message, and the content is a message sent to a user.

6. An information processing method comprising: a step of extracting first BCT component information relating to components of a first behavior change technique (first BCT) contained in first content determined to be subject to filtering; a step of extracting second BCT component information relating to components of a second behavior change technique (second BCT) contained in second content not determined to be subject to filtering; and a step of estimating reference information that serves as a criterion for determining whether or not a content is subject to filtering based on the first BCT component information and the second BCT component information, and generating a model based on the reference information.

7. An information processing program for causing a computer to execute the following steps: extracting first BCT component information regarding components of a first behavior change technique (first BCT) contained in first content determined to be subject to filtering; extracting second BCT component information regarding components of a second behavior change technique (second BCT) contained in second content not determined to be subject to filtering; and estimating reference information that serves as a criterion for determining whether or not a content is subject to filtering based on the first BCT component information and the second BCT component information, and generating a model based on the reference information.

Images