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

Abstract

To improve precision of filtering of content.SOLUTION: An information processing system 1 includes a first extraction unit 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 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 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.SELECTED DRAWING: Figure 2B

Description

[Technical field]

[0001] The present invention relates to an information processing system, an information processing method, and an information processing program. [Background technology]

[0002] It has been known to filter various types of information using models based on machine learning. For example, in Patent Document 1, learning data for a Bayesian filter is converted into a learning sequence. The system manages the messages as a learning spam filter and uses the learning spam filter to identify unwanted messages. It is stated that: [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Patent No. 5676709 Summary of the Invention [Problem to be solved by the invention]

[0004] However, for example, in the conventional spam filter described in Patent Document 1, may mistakenly identify non-spam messages as spam (false positives), There is a risk that spam messages may accidentally slip through the filter (false negatives). The problem arises not only with spam filters, but also with filters of any kind of content. There is a risk.

[0005] Therefore, an information processing system and an information processing method capable of improving the accuracy of content filtering are provided. One of the objects of the present invention is to provide a method and an information processing program. [Means for solving the problem]

[0006] According to an aspect of the present invention, there is provided an information processing system for detecting a first item determined to be a filtering target. 1. BCT components of the 1st behavior change technique (1st BCT) included in 1 content The first extraction unit extracts information, and the second content extraction unit extracts information that is not determined to be subject to filtering. The second BCT component information was extracted from the second BCT component information included in the content. and a second extraction unit that outputs, based on the first BCT component information and the second BCT component information, Estimate criteria information that is a criterion for determining whether or not the content is subject to filtering, and A generating unit that generates a model based on the quasi-information.

[0007] According to another aspect of the present invention, there is provided an information processing method, in which an information processing device is a filtering target. Regarding the components of the first behavior change technique (first BCT) included in the first content that was judged to be A step of extracting first BCT component information and a step of extracting second BCT component information that is not determined to be a filtering target. 2nd BCT components related to the components of the 2nd behavior change technique (2nd BCT) included in the 2nd content extracting information based on the first BCT component information and the second BCT component information; The method estimates reference information that is a criterion for determining whether or not the content is a filtering target, and and generating a model based on the reference information.

[0008] An information processing program according to another aspect of the present invention includes: The components of the first behavior change technique (first BCT) included in the first content that was determined to be A step of extracting first BCT component information related to the filtering target The second BCT is a component of the second behavior change technique (second BCT) included in the second content. A step of extracting T component information, and Based on this, we estimate the criteria for determining whether or not the content is subject to filtering. and generating a model based on the reference information. Effect of the Invention

[0009] According to the present invention, the accuracy of content filtering can be improved. [Brief description of the drawings]

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

[0011] An embodiment of the present invention will be described with reference to the accompanying drawings. Those marked with the same symbol have the same or similar configuration.

[0012] (overview) In the information processing system according to this embodiment, the information processing device The components of the first behavior change technique (first BCT) included in the first content that was determined to be The information processing device extracts information related to the filter (hereinafter referred to as "first BCT component information"). The second behavior change technique ( Information regarding the components of the first BCT (hereinafter referred to as "second BCT component information") is extracted. The information processing device performs filtering based on the first BCT component information and the second BCT component information. Information that serves as the criteria for determining whether or not content is subject to filtering (hereinafter referred to as "criteria information") The extraction is performed by estimating the standard information, and a model is generated based on the standard information. This can also be rephrased.

[0013] In this embodiment, the behavior change technique (BCT) is a method for It is a method or technique that affects (influences) dynamic transformation. 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 f or the reporting of behavior change interventions. Ann Behav Med 2013; 46: 81~9 5.) specifies 93 BCTs in 16 groups. As long as it comprehensively covers behavioral change methods, not limited to CTTv1, how is it stipulated? In addition, the BCT according to this embodiment may be a group including one or more BCTs (hereinafter, “B The BCT group and BCT may be used interchangeably. stomach.

[0014] FIG. 1 is a diagram showing an example of a classification of BCTs according to the present embodiment. One or more BCTs are As shown in Figure 1, BCTTv1 includes the following categories: planning”, “2.Feedback and monitoring”, “3.Social support”, “4.Shapi ng knowledge”, “5. Natural consequences”, “6. Comparison of behavior”, “7.Association”, “8.Repetition and substitution”, “9.Comparison of o ``10.Reward and threat'', ``11.Regulation'', ``12.Antecedent s”, “13.Identity”, “14.Scheduled consequences”, “15.Self-belief” In addition, 16 BCT groups called "16. Covert learning" have been defined.

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

[0016] In addition, each BCT has a component, and the degree to which each BCT is included in a certain content is expressed as a component value. Also, the sum of the component values ​​of each BCT belonging to the same BCT group is the BCT It may be shown as a component value of the group.

[0017] 2A and 2B show the generation of a model for filtering according to the conventional technique and the present embodiment. In FIG. 2B, the content to be filtered is spam messages. The first message that is determined to be spam and / or junk and is subject to filtering The second content is determined to be spam and not subject to filtering. Messages whose content is not spam (hereinafter referred to as "normal messages") However, FIG. 2B is merely an example, and the content to be filtered may be The first content is not limited to spam messages, and the second content is a normal message. Not limited to.

[0018] Here, a spam message is one that is intended to give a user (e.g., the message reader) good intentions. Exaggerated the costs of not acting quickly without providing any basis for decision-making By placing psychological burden on users with statements that incite feelings of inferiority, it is difficult to maintain a healthy user attitude. These are messages that are deliberately designed to interfere with or hinder your decision. For example, emails, short messages, and sites called "How was your experience?" Any information disseminated through various media, including stealth marketing sites, is acceptable. In addition, spam messages are sent in the form of at least text data, image data, and audio data. It may be just one.

[0019] In addition, spam messages can be said to be information that contains an excessive amount of specific BCT components. The BCTs in this category are, for example, BCTs that belong to at least one of the following categories: sludge, threats, and incitement ( For example, in Figure 1, "5.5. Anticipated regret" and / or "10.11. Future punis Sludge is a type of fraud that companies or fraud groups take advantage of for their own profits. It may induce actions that are unfavorable to the user or hinder actions that are favorable to the user. For example, in the context of spam messages, sludge refers to the act of sending credit to fraudulent sites. They may ask you to enter your credit card number or make a cash transfer for a fictitious charge. This may include not giving the person sufficient time to consult with their family before committing a crime.

[0020] For example, as shown in FIG. 2A, in one example of the prior art, a message is annotated as a nuisance message. Machine learning is performed using the training data collected by the algorithm to generate a decision criterion based on Bayesian estimation. A model for filtering based on the judgment criteria is generated. The criterion is to estimate the probability of a message being spam using conditional probability and a predefined threshold. If the message exceeds this threshold, it is determined to be spam. " may be rephrased as tagging or labeling, etc.

[0021] Conventional filtering techniques based on training data annotated as spam messages Other well-known techniques include neural networks and random forests. Lists that you have set up for your application (e.g., blacklists, real-time blacklists, reputation lists, etc.) Blocking techniques based on block lists, etc. are also known. As spam messages become more individualized and sophisticated, the above-mentioned filtering and blocking technologies According to the model using, the probability of false positives and false negatives increases, and the accuracy of filtering decreases. There is a risk that this may happen.

[0022] On the other hand, as shown in FIG. 2B, in the information processing system according to the present embodiment, the nuisance message The first BCT components included in the first training data annotated as The CT component information is extracted and annotated as a normal message in the second training data. Second BCT component information regarding the included second BCT components is extracted. By analyzing the difference between the second BCT component information and the second BCT component information, it is possible to determine whether the message is a spam message or not. Criteria information that will be used as a criterion for judgment is estimated, and the filtering of unwanted messages is performed based on the criteria information. A model for the analysis is generated.

[0023] In this way, in the information processing system according to this embodiment, The first content that was rejected (e.g., the first teacher annotated as a spam message) The first BCT component information was extracted from the data and determined not to be subject to filtering. Secondary content (e.g., secondary training data annotated as normal messages) The second BCT component information is extracted from the first BCT component information and the second BCT component information. Based on the comparison of CT component information, the content to be filtered (e.g., spam messages) The reference information is estimated as a criterion for determining whether or not a certain item is a page, and a model based on the reference information is created. Therefore, compared to conventional standards, the probability of false positives and false negatives is reduced. It is possible to generate criteria-based models, which can improve the accuracy of filtering. do.

[0024] (composition) Next, a configuration of the information processing system 1 according to the present embodiment will be described. The system 1 may be configured to include a single information processing device, or may include multiple information processing devices. For example, the information processing system 1 may be configured to implement a part of the functional configuration described later. It is composed of multiple information processing devices and is connected by a wired and / or wireless network. The multiple devices may implement the functional configuration described below.

[0025] FIG. 3 shows hardware of each information processing device constituting the information processing system according to this embodiment. FIG. 1 is a diagram illustrating an example of a configuration of an information processing system 1. The generating device 10 and the filter device 20 are each a central processing unit (CPU) that corresponds to a calculation device. A processor 11 such as a processing unit (CPU), a storage device 12, a communication device 13, and an input / output device 14. These components are connected to each other via a bus so as to be able to transmit and receive data.

[0026] The processor 11 is, for example, a CPU, and controls each information processing device in the information processing system 1. The processor 11 reads out a program from the storage device 12 and executes it. Each information processing unit in the information processing system 1 may execute various processes described in the present embodiment. The processing device may include one or more processors 11. It may also be called a puter.

[0027] The storage device 12 may be, for example, a memory, a hard disk drive (HDD), and / or a storage device (SSD). The storage device 12 is configured by the processor 11. Various information required for executing the processing (for example, the information processing program executed by the processor 11) The device may also store the following information:

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

[0029] The input / output device 14 is, for example, a keyboard, a touch panel, a mouse, and / or a microphone. The input unit includes an input section and an output section, such as a display and / or a speaker. The output unit receives various information from the user. The device may include a display unit that displays the

[0030] The above-described hardware configuration is merely an example. The device may be configured such that some of the hardware shown in FIG. 4 is omitted, or the device may be configured such that the hardware shown in FIG. In addition, the present invention may include one or more of the hardware shown in FIG. In addition, when the information processing device is composed of a plurality of devices, In this case, each device may include at least some of the hardware.

[0031] FIG. 4 is a diagram showing an example of a functional configuration of an information processing system according to the present embodiment. As shown, the information processing system 1 is an information processing device that generates the above model (for example, FIG. 2B). A generating device (hereinafter referred to as a generating device) 10 acquires content and assigns the acquired content to a An information processing device (hereinafter, a filter device) 20 that performs filtering based on the model; The generating device 10 and the filter device 20 are configured as a single information processing device. The information processing device may be configured as a single device, or may be configured as a separate information processing device.

[0032] 4 is merely an example, and the generating device 10 and the filter device 20 may be implemented by other devices (not shown). Of course, the functions of each functional configuration may be implemented by the processor 1 shown in FIG. By loading an information processing program into the hardware such as processor 11, At least one of the storage device 12, the communication device 13, and the input / output device 14 is operated under control. This is achieved by

[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 03 and a generation unit 104.

[0034] The storage unit 101 stores the first content determined to be a filtering target and the The storage unit stores the second content that is not determined to be a ring target. 101 is a first training data annotated as a filtering target (e.g., The training data (annotated as spam messages) is stored as the first content. In addition, the storage unit 101 may store the information that is annotated as not being subject to filtering. The second training data (e.g., training data annotated as normal messages) is used as the first training data. It may be stored as 2 contents.

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

[0036] The first extraction unit 102 extracts the first content determined to be a filtering target. Specifically, the first extraction unit 1 extracts first BCT component information related to the components of the first BCT. 02 is the first content (e.g., the content annotated as a filtering target) The first BCT component information may be extracted by machine learning using the first teacher data.

[0037] The first BCT component information is the average of the first BCT components contained in the first content. Alternatively, the first content may be included in the first content. The first BCT component information may indicate a distribution of the components of the first content. Information regarding the ratio of the amount of information of the component of the first BCT to the amount of information of the component of the first BCT (hereinafter, "the first "Content information") and a specific BCT for the amount of information corresponding to the component of the first BCT At least information on the ratio of the amount of information of the components of the CT (hereinafter referred to as "first composition ratio information") It may also include one.

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

[0039] For example, in FIG. 5A, the first BCT component information is annotated as spam. If the amount of information in the first training data is 100%, the amount of information in the threat-based BCT component is 50%. The information content of the inciting BCT components is 40%, and the information content of the other BCT components is 5%. This indicates that the amount of information that does not belong to any BCT (no BCT) is 5%. In this case, the first BCT (here, the BCT of the threatening type and the BCT of the provocation type) included in the first teacher data is The amount of information on the components of BCT and other BCT is 95%, so the first content information is 95 In addition, the information content of the first BCT component is 95% and the information content of the specific BC The information content of the T component (here, the threatening BCT and the provocative BCT) is 90% (=40% + 50%), the first component ratio information may indicate 90 / 95.

[0040] The second extraction unit 103 extracts the second content that is not determined to be a filtering target. The second BCT component information related to the included second BCT components is extracted. The output unit 103 outputs the second content (for example, The second BCT component information may be extracted by machine learning using the second training data (the second training data obtained by the above process). .

[0041] The second BCT component information is the average of the second BCT components included in the second content. Alternatively, the second content may include a second content table showing a fixed value based on the second content. The second BCT component information may indicate a distribution of the components of the second content. Information regarding the ratio of the amount of information of the second BCT component to the amount of information of the second BCT component (hereinafter, "the second "Content rate information"), and a specific BCT for the amount of information corresponding to the component of the second BCT At least information on the ratio of the amount of information of the components of the CT (hereinafter referred to as "second composition ratio information") It may also include one.

[0042] FIG. 5B is a diagram showing an example of second BCT component information according to this embodiment. For example, FIG. In B, the second training data is annotated as a normal message. An example of BCT component information is shown. Also, in FIG. 5B, the second BCT component information is It is a component table that shows fixed values ​​based on the average of the components of the training data, as described above. The distribution of the components of the second training data may be shown.

[0043] For example, in FIG. 5B, the second BCT component information is annotated as a normal email. If the amount of information in the second training data is 100%, it does not belong to any BCT (no BCT ) information content is 70%, the information content of the comparison BCT is 10%, and the information content of the planning BCT is The information content of the second BCT is 10%, and the information content of other BCTs is 10%. The second BCT included in the teacher data (here, the comparison BCT, the planning BCT, and other The amount of information corresponding to the component of BCT is 30%, so the second content rate information is 30 / 10 0. Also, the amount of information corresponding to the second BCT component is 30%, and a specific B The amount of information corresponding to the components of CT (here, the BCT of the threatening type and the BCT of the provocation type) is 0%. Therefore, the second component ratio information may indicate 0.

[0044] The generation unit 104 generates a filtering function based on the first BCT component information and the second BCT component information. The method estimates reference information that is a criterion for determining whether the content is a target for downloading, and performs a process based on the reference information. The model is generated based on the criteria information, which is based on the amount of information in the content to be filtered. The ratio of information content of the third behavior change technique (third BCT) included in the content Information on the amount of information on the third BCT components (hereinafter referred to as "content information"), Information on the ratio of the information content of the BCT components (hereinafter referred to as "composition ratio information") and Including at least one piece of information regarding the specific BCT (hereinafter referred to as "BCT information"). Good too.

[0045] Specifically, the generation unit 104 generates a filter based on the first content rate information and the second content rate information. The amount of information in the content to be tagged is compared to the amount of 3rd BCT components in the content For example, the content ratio information of the information content ratio of the As shown above, the first content rate information is 95 / 100 and the second content rate information is 30 / 100. In this case, the generation unit 104 generates a filtering target content (i.e., a spam message). The content rate information that serves as the criterion for determining whether or not the substance is a substance may be estimated to be 50 / 100.

[0046] In addition, the generation unit 104 generates the third BCT based on the first composition ratio information and the second composition ratio information. Estimate the composition ratio information regarding the ratio of the information content of a specific BCT component to the information content of the component For example, as described with reference to FIGS. 5A and 5B, when the first composition ratio information is 90 / 95, If the second component ratio information is 0, the generation unit 104 generates the content to be filtered ( In other words, the composition ratio information that is the criterion for determining whether a message is spam is estimated to be 40 / 50. This is also fine.

[0047] Furthermore, the generating unit 104 generates a specific BCT component based on the first BCT component information and the second BCT component information. For example, as shown in FIGS. 5A and 5B, The first BCT component information shows that the ratio of the threatening BCT and inciting BCT components is higher than that of the normal message. This is higher than the proportion of the components of the comparison system BCT and the planning system BCT that are included in the message. 4, based on a comparison of the first BCT component information and the second BCT component information, a filtering target is determined. Identify specific BCTs that tend to be included in elephant content (i.e., spam messages) BCT information may be estimated.

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

[0049] The acquisition unit 201 acquires the content. Specifically, the acquisition unit 201 performs filtering. Content that is subject to scrutiny (e.g., whether it is a spam message or not) Obtain the target message that will be used to judge whether the

[0050] The filter unit 202 performs the filtering process based on the model generated by the generation device 10. It is determined whether the acquired content is subject to filtering or not, The filter unit 202 performs filtering based on the result of the determination. The model generated by the above method may be used as it is, or a model obtained by distilling the model may be used. As described above, the model may determine at least one of the content rate information and the composition ratio information. It may be used as a cutoff criterion.

[0051] Specifically, the filter unit 202 filters the information of the content acquired by the acquisition unit 201. Based on the ratio of the amount of BCT information contained in the content to the amount of The model may be used to determine whether the content is subject to filtering. For example, in this model, the corresponding The ratio of the amount of BCT information contained in the content is the value indicated by the content rate information above (for example, 50 / 100) or how much it deviates from the value indicated by the content rate information. In this manner, it may be possible to determine whether the content is subject to filtering.

[0052] In addition, the filter unit 202 filters B included in the content acquired by the acquisition unit 201. The ratio of the information content of a specific BCT component to the information content of the CT and the model based on the above composition ratio information The controller may be used to determine whether the content is subject to filtering. In this model, the information of the BCT included in the content acquired by the acquisition unit 201 is The ratio of a specific BCT component to the amount is the value indicated by the composition ratio information above (e.g., 40 / 50 ) or how much it deviates from the value indicated by the composition ratio information. It may be determined whether the content is subject to filtering.

[0053] In addition, the filter unit 202 filters B included in the content acquired by the acquisition unit 201. The ratio of the information amount of a specific BCT component to the information amount of the CT, and the above content rate information and composition ratio Uses an information-based model to determine whether content is eligible for filtering Good too.

[0054] The output unit 203 outputs information on the result of the filtering by the filter unit 202 (hereinafter, The filtering information is output from the generating device 1. For example, the filtering unit 202 may feed back the filtered signal to 0. Content that was previously filtered as illegal content is now exempt from filtering. If the content is not appropriate for the purpose (i.e., if it is a false positive), the filtered content The content is annotated as not subject to filtering and generated as the second training data. The filter unit 202 may then feed back the signal to the device 10. Content that was not filtered as content to be filtered is If the content is something that should be targeted (i.e., a false negative), The content that was not included in the filter is annotated as a second training data. The generated data may be fed back to the generating device 10. This allows for further improvement in the accuracy of filtering based on the model used. (operation) Next, the operation of the information processing system 1 according to the present embodiment will be described. The operation of the generating device 10 and the filter device 20 shown is merely an example and is not limited to what is shown. For example, some steps may be deleted and / or added, and some steps (e.g. For example, the order of steps S101 and S102 in FIG. 6 may be reversed. may be executed in parallel.

[0055] FIG. 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 generates a first content that is determined to be a filtering target. The first BCT component information regarding the components of the first BCT included in the content is extracted (step S1 01). For example, as described in FIG. 2B and FIG. 5A, the generating device 10 may The first BCT component information is extracted by machine learning using the annotated first training data. You may put it out.

[0056] The generating device 10 detects the content included in the second content that is not determined to be a filtering target. Then, second BCT component information relating to the components of the second BCT included in the second BCT is extracted (step S102). For example, as described in FIGS. 2B and 5B, the generating device 10 may generate a mail that is annotated as a normal mail. The second BCT component information can be extracted by machine learning using the second training data. good.

[0057] The generating device 10 extracts the first BCT component information in step S101 and Based on the second BCT component information extracted in 02, the content to be filtered is The reference information that is the criterion for determining whether or not the object exists is estimated (step S103). As described in 2B, 5A and 5B, the generating device 10 generates the first BCT component information and the second BCT component information. The difference in the CT component information may be analyzed to estimate the reference information that serves as the judgment criterion. The apparatus 10 generates a model based on the reference information estimated in step S103 (step S104).

[0058] FIG. 7 is a flowchart showing an example of the operation of the filter device 20 according to the present embodiment. As shown in FIG. 7, the filter device 20 determines whether a signal is a filtering target. For example, the filter device 20 obtains the content that is a nuisance. A target message to be judged as a message or not is obtained.

[0059] The filter device 20 performs the following steps based on the model generated by the generation device 10: It is judged whether the content acquired by is a content to be filtered or not, and Filtering is performed based on the result of the analysis (step S202). For example, the filter device 2 0 judges whether the target message is a spam message based on the model. , and filtering is performed based on the judgment results.

[0060] The filter device 20 performs a filter operation on the result of the filtering in step S202. Then, the controller 11 outputs the tracking information (step S203).

[0061] As described above, in the information processing system 1 according to the present embodiment, the generating device 10 The first BCT component information extracted from the first content determined to be a ring target and the Second BCT component information extracted from the second content that is determined not to be filtered Based on the information, it is possible to determine whether the content is subject to filtering (e.g., spam messages). Reference information is estimated as a criterion for making such a judgment, and a model based on the reference information is generated. Therefore, compared with conventional criteria, the filtering criteria have a lower probability of false positives and false negatives. It can generate models for filtering.

[0062] In the information processing system 1 according to the present embodiment, the above-mentioned Based on the above model, the retrieved content is checked to see if it is a content to be filtered. Then, based on the result of the judgment, filtering is performed. This can improve the accuracy of the analysis.

[0063] (Other embodiments) In the above embodiment, a spam message is exemplified as a content to be filtered. However, the content to be filtered is not limited to the above. The content is designed by, for example, catchphrases, advertisements, and prompt engineering. The first content that is likely to contain a BCT showing a remarkable tendency, such as a content If there is a combination of secondary contents with a low probability of including the BCT showing the trend, It is possible to filter such content. For example, the first content The combination of the first and second content is affected by the quality of the catch phrase, advertising The balance of the two is good or bad, or the content is designed by Prompt Engineering. The content may be determined based on whether it is appropriate or inappropriate.

[0064] In addition, the content to be filtered is "content that encourages specific actions." For example, the generating device 10 may send a coercive email to a person in a weaker position than the generating device 10. A model is generated to judge whether the user is trying to encourage a behavior or not, and the filter device uses the model Based on this, the target content is a coercive email to encourage action by a person in a weaker position. Such models may be judged as being sexually harassing. ), power harassment, and bullying of subcontractors. .

[0065] In addition, certain BCTs may not be able to filter content that is determined to be subject to filtering. BCT that is expected to contain more significant content than content that is determined not to be subject to tagging. Any BCT can be used as long as it is well written. For example, When filtering behavior, a specific BCT may provide an objective view of the behavior, for example. However, it is not limited to this. In addition, it is possible to use BCTs that are judged to be well-balanced. When filtering content that may be used, a specific BCT may, for example, The BCT may be, but is not limited to, a BCT that presents the It is designed by the ML and filters content that is deemed appropriate. In this case, the specific BCT may be, for example, a BCT of a lineage that provides feedback. But not limited to this.

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

[0067] The above-described embodiment is intended to facilitate understanding of the present invention and does not limit the present invention. The flowcharts, sequences, and implementations described in the embodiments are not intended to be interpreted as being related to the present invention. The elements of the embodiment, as well as their arrangement, materials, conditions, shapes, sizes, etc. are merely examples. It is not limited to this and can be changed as appropriate. [Explanation of symbols]

[0068] 1...information processing system, 10...generation device, 20...filter device, 11...processor, 1 2...storage device, 13...communication device, 14...input / output device, 101...storage section, 102...first extraction unit, 103... second extraction unit, 104... generation unit, 201... acquisition unit, 202... filter unit, 20 3. Output section

Claims

1. A first behavior modification technique ( A first extraction unit that extracts first BCT component information related to a component of a first BCT; A second behavior change included in a second content that is not determined to be subject to filtering a second extractor that extracts second BCT component information related to components of a method (second BCT); Based on the first BCT component information and the second BCT component information, The method estimates reference information that is a criterion for determining whether the content is a target content or not, and creates a model based on the reference information. A generator for generating a rule; An information processing system comprising:

2. The reference information is a ratio of an amount of information of the content to be filtered. Content rate information regarding the ratio of the information amount of the component of the third behavior change technique (third BCT) included in A configuration regarding a ratio of the information amount of a specific BCT component to the information amount of the third BCT component. ratio information; and BCT information relating to the specific BCT.

2. The information processing system according to claim 1.

3. The first BCT component information is a result of the first BCT with respect to an amount of information of the first content. First content ratio information on the ratio of information content of the first BCT and information corresponding to the first BCT component At least one of the first composition ratio information regarding the ratio of the information amount of the specific BCT component to the amount of the specific BCT component. Including one The second BCT component information is a result of the second BCT with respect to the amount of information of the first content. Second content ratio information relating to the ratio of information content of the second BCT and information corresponding to the component of the second BCT At least one of the second composition ratio information regarding the ratio of the information amount of the specific BCT component to the amount of the specific BCT component. Including one 3. The information processing system according to claim 2.

4. An acquisition unit that acquires content; Based on the model, the acquired content is filtered. a filter unit that judges whether the content is a content or not and performs filtering based on the judgment result; The information processing system according to claim 1 , further comprising:

5. the first content is an unsolicited message; the second content is a normal message other than a spam message, The content is a message sent to a user.

5. The information processing system according to claim 4.

6. An information processing device, A first behavior modification technique ( extracting first BCT component information relating to components of the first BCT; A second behavior change included in a second content that is not determined to be subject to filtering Extracting second BCT component information relating to components of the method (second BCT); Based on the first BCT component information and the second BCT component information, The method estimates reference information that is a criterion for determining whether the content is a target content or not, and creates a model based on the reference information. and An information processing method comprising the steps of:

7. On the computer, A first behavior modification technique ( extracting first BCT component information relating to components of the first BCT; A second behavior change included in a second content that is not determined to be subject to filtering Extracting second BCT component information relating to components of the method (second BCT); Based on the first BCT component information and the second BCT component information, The method estimates reference information that is a criterion for determining whether the content is a target content or not, and creates a model based on the reference information. and An information processing program for executing the above.

Images