Information processing device

Abstract

The objective of the present invention is to provide a time-series data processing device that is able to reduce the load when time-series data such as user behavior is handled in a machine learning process.　A behavior analysis device 100 is provided with a time-series data acquisition unit 101 for acquiring noncritical behavior data among one or a plurality of time-series data representing user behavior, and a generalization processing unit 102 for subjecting the noncritical behavior data to a generalization process. In the present disclosure the noncritical behavior data indicates behavior having a low frequency among the behavior history of a user. As a generalization process, the generalization processing unit 102 replaces the noncritical behavior data with a prescribed character string (for example, [UNK]). In addition, the generalization processing unit 102 subjects the noncritical behavior data to the generalization process so as to indicate the relevant behavior category.

Description

[Technical Field]

[0001] The present invention relates to an information processing device that processes time-series data indicating user behavior and the like. [Background technology]

[0002] A known machine learning method for natural language processing is BERT (Bidirectional Encoder Representations from Transformers), which is described in Non-Patent Document 1. This BERT has an encoder-decoder model with a self-attention mechanism, and performs natural language processing and image processing. [Prior art documents] [Non-patent literature]

[0003] [Non-Patent Document 1] Jacob Devlin and 2 others, BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding, [online], Posted: May 24, 2019, Published by: arXiv.org, Retrieved: July 2, 2021, https: / / arxiv.org / pdf / 1810.04805.pdf, English Summary of the Invention [Problem to be solved by the invention]

[0004] The passage of time in each customer experience or user behavior is important for understanding that customer experience or user behavior. BERT must be able to properly learn the adjacency relationships between local and global time-series data. However, while the passage of time in each customer experience or user behavior is important for understanding that customer experience or user behavior, distinguishing between user behaviors that occurred at different times and learning them would require a massive learning effort.

[0005] Therefore, in order to solve the above-mentioned problems, an object of the present invention is to provide an information processing device that can reduce the load when handling behavioral data such as user behavior in machine learning processing. [Means for solving the problem]

[0006] The information processing device of the present invention includes an acquisition unit that acquires non-important behavior data from one or more behavior data indicating user behavior, and a processing unit that generalizes the non-important behavior data. [Effects of the Invention]

[0007] According to the present invention, data can be generated that reduces the load when performing machine learning processing. [Brief explanation of the drawings]

[0008] [Figure 1] 1 is a system configuration diagram of a communication system including a behavior analysis device 100 that acquires and analyzes user behavior according to the present disclosure. [Figure 2] FIG. 2 is a block diagram showing the functional configuration of the behavior analysis apparatus 100. [Figure 3] FIG. 10 is a schematic diagram showing a specific example of generalization processing. [Figure 4] FIG. 10 is a schematic diagram showing a behavior identifier insertion process. [Figure 5] FIG. 1 is a diagram illustrating a model that schematically illustrates a BERT. [Figure 6] 10 is a flowchart showing the learning process of behavior analysis apparatus 100. [Figure 7] 10 is a flowchart showing the processing performed by behavior analysis apparatus 100. [Figure 8] FIG. 2 is a diagram showing a behavior history database 101a. [Figure 9] This is a diagram showing the frequency of occurrence of each category for all users. [Figure 10] This is a diagram in which the frequency of user behavior classifications is tallied for each classification organized by hierarchy. [Figure 11] FIG. 10 is a diagram in which the summary tables shown in FIG. 9 and FIG. 10(a) to FIG. 10(c) are merged. [Figure 12] This is a diagram showing the top 12 categories by frequency of occurrence extracted from the merged summary table. [Figure 13] FIG. 13 is a diagram showing generalized behavior identifiers based on the management table (FIG. 12). [Figure 14] FIG. 10 is a diagram in which behavior identifiers indicating divisions based on time intervals between behaviors are inserted into generalized behavior history data. [Figure 15] FIG. 10 is a diagram illustrating a process of calculating and selecting attention weights from a user behavior history database 101a. [Figure 16] 1 is a diagram illustrating an example of a hardware configuration of a behavior analysis apparatus 100 according to an embodiment of the present disclosure. DETAILED DESCRIPTION OF THE INVENTION

[0009] The present disclosure will be described with reference to the accompanying drawings. Whenever possible, the same parts are designated by the same reference numerals and redundant description will be omitted.

[0010] 1 is a system configuration diagram of a communication system including a behavior analysis device 100 according to the present disclosure that acquires and analyzes user behavior. This behavior analysis device 100 functions as a time-series data processing device that handles the operation history of a user's access to a website as time-series data. Furthermore, when a user uses a company's telephone service (such as for product consultations or questions), the behavior analysis device 100 handles the usage status as time-series data.

[0011] Behavior analysis device 100 collects and analyzes the access history of PC 200 operated by the user to be analyzed to web server 300. In addition, the behavior history of users who have used telephone services is recorded by the service administrator and registered in behavior history database 101a of behavior analysis device 100.

[0012] The PC 200 is a general personal computer, and accesses the web server 300 via a network.

[0013] The web server 300 is a server that provides web information to the PC 200 .

[0014] Behavior analysis device 100 collects behavioral history data on what information on web server 300 PC 200 has accessed. For example, when PC 200 accesses the site of a mobile phone carrier, it collects what information was accessed. More specifically, it collects information on the mobile phone rate plan and the type of mobile terminal.

[0015] 2 is a block diagram showing the functional configuration of behavior analysis device 100. This behavior analysis device 100 includes a behavior history database 101a, a time-series data acquisition unit 101, a generalization processing unit 102, a data processing unit 103, a learning unit 104, a learning model 104a, a BERT processing unit 105, a range selection unit 106, and an output unit 107. Each component will be described below.

[0016] The behavior history database 101a is a section that stores behavior history data when the PC 200 accesses the web server 300. The PC 200 or the web server 300 stores the behavior history data in the behavior history database 101a each time it is accessed or periodically. As described above, the behavior history database 101a also stores user behavior other than web accesses in accordance with operations by an operator. In the present disclosure, when a user receives a service equivalent to a service provided by the web server 300 by a method other than the web server 300, for example, by telephone, the user behavior is registered by a telephone operator or the like. In the present disclosure, user behavior refers to the behavior of a user in receiving a certain service when the user attempts to receive that service.

[0017] The time-series data acquisition unit 101 is a part that acquires time-series data, which is behavior history data, from the behavior history database 101a.

[0018] The generalization processing unit 102 generalizes time-series data that occur infrequently among the time-series data. For example, as the generalization processing, the generalization processing unit 102 replaces all or part of the time-series data that occur infrequently with a predetermined symbol or character string.

[0019] FIG. 3 is a schematic diagram showing a specific example of the generalization process. In the figure, behavior identifiers indicated by behavior A1, behavior A2, etc. are shown as access information for a website. In FIG. 3(a), a certain user performed behavior A1, behavior A2, etc., behavior B3, behavior B4, and behavior B5. The behaviors A1, etc. indicate behavior identifiers and indicate access to a certain piece of information on the web server 300. Furthermore, behavior A1 and behavior B1 indicate different behavior categories. The generalization processing unit 102 then replaces low-frequency behaviors with behavior category + [UNK]. For example, if the occurrence frequency of behavior B4 is low, the generalization processing unit 102 replaces behavior B4 with B[UNK] (FIG. 3(b)). Further detailed processing by the generalization processing unit 102 will be described later.

[0020] The data processing unit 103 is a part that inserts a behavior identifier for distinguishing between time series data indicating a plurality of user behaviors when there is a predetermined time interval between the time series data.

[0021] FIG. 4 is a schematic diagram showing the insertion process of an action identifier. Note that in this diagram, the notation of categories is omitted. As shown in FIG. 4(a), it is assumed that there is 6 minutes between action 1 and action 2, and 26 minutes between action 2 and action 3. It is assumed that other actions are also as shown in the diagram. In such user actions, there are often cases where the actions are continuous in time and there is no clear division. Furthermore, the time intervals between user actions often vary.

[0022] As shown in Figure 4(b), the data processing unit 103 inserts behavior identifiers representing divisions equal to the maximum integer number equal to or less than the real number obtained by taking the logarithm of the interval between behaviors with the threshold parameter. For example, as shown in Figure 4, the interval between behavior 1 and behavior 2 is 6 minutes. If the threshold parameter is 5, the data processing unit 103 calculates Log5(6) = 1 and inserts one behavior identifier representing a division between behavior 1 and behavior 2, which is an interval of 6 minutes. Similar calculations are performed for other behaviors.

[0023] By inserting behavioral identifiers that indicate breaks in this way, it is possible to indicate breaks in behavior. Furthermore, using a so-called log scale, the longer the time interval, the more behavioral identifiers that indicate breaks are inserted, and a function is used to adjust the number to limit it when the time interval becomes longer. This makes it possible to show the relationship between the length of the time interval and the number of behavioral identifiers that represent breaks. Depending on the user's behavior, the meaning of the behavior often does not change after a certain amount of time has passed. For example, a user's behavior performed five minutes later must be clearly distinguished from a user's behavior performed one year later, but there may be cases where it is not necessary to distinguish between the meaning or content of a user's behavior performed one year later and a user's behavior performed two years later.

[0024] Note that instead of or in addition to inserting an activity identifier indicating a division, the data processing unit 103 may insert an activity identifier according to the length of the user's activity. For example, when the user activity includes being at home, going out, or traveling, the data processing unit 103 may add an activity identifier according to the length of the time spent at home, the time spent outside, or the time spent traveling. For example, the activity identifier indicating being at home may be added so that one activity identifier indicates one hour, and two activity identifiers indicate two hours. Furthermore, an activity identifier indicating time may be added either before or after the activity identifier.

[0025] The learning unit 104 is a part that performs machine learning using time-series data that has been generalized and processed. In the present disclosure, machine learning is performed using BERT (Bidirectional Encoder Representations from Transformers), which is used in language models for natural language processing. In BERT, learning processing is performed by performing pre-learning and fine tuning. In pre-learning, fill-in-the-blank processing and adjacent prediction processing are performed using time-series data. In normal BERT learning, sentences are input, but in the present disclosure, as described above, time-series data that has been generalized and processed is input to generate a learning model 104a using BERT.

[0026] FIG. 5 is a diagram showing a model that schematically illustrates BERT. This model receives a sequence of vectors (E1...EN) as input and outputs a sequence of vectors (T1...TN). The model also has a structure in which multiple hidden layers (Trm = Transformer) are stacked between the input layer and the output layer. This diagram is an excerpt from Non-Patent Document 1.

[0027] The BERT processing unit 105 is a part that performs processing using a BERT-based learning model 104a. The BERT processing unit 105 uses a self-attention function in the BERT learning model 104a to calculate attention weights that indicate the degree of correlation between multiple pieces of input time-series data. The learning unit 104 learns the learning model 104a to calculate these attention weights.

[0028] The range selection unit 106 is a part that derives behavior related to a specified user behavior based on the attention weight calculated by the BERT processing unit 105. The range selection unit 106 accepts time series data to be compared, compares the attention weight between the time series data to be compared with a pre-input threshold, and selects time series data having an attention weight equal to or greater than the threshold. Note that the range selection unit 106 may select time series data from the oldest time series data onward. In this case, time series data with an attention weight less than the threshold may be included.

[0029] The output unit 107 is a part that outputs the selected time series data. In the present disclosure, the output unit 107 outputs the data to a display unit or to the outside via a communication unit.

[0030] Next, a description will be given of the operation of behavior analysis apparatus 100 of the present disclosure.

[0031] The time-series data acquiring unit 101 receives behavior history data of multiple users and a time interval threshold parameter (S101). Then, the time-series data acquiring unit 101 acquires the date and time of the behavior, the user identifier, and the behavior identifier from the behavior history data (S102).

[0032] The generalization processing unit 102 sorts the behavioral identifiers acquired for each user based on the user identifier, and performs generalization processing on the behavioral identifiers with low occurrence frequency among the behavioral identifiers for each user, i.e., replaces the behavioral identifiers with low occurrence frequency with generalized symbols (S103).

[0033] The data processing unit 103 inserts behavior identifiers indicating breaks between the behavior identifiers based on the threshold parameter (S104). These processes are performed on, for example, 1000 pieces of time-series data.

[0034] The learning unit 104 performs a learning process using BERT, generates a learning model 104a using BERT, and stores it (S105, S106). For example, the learning process is performed using 1000 pieces of time-series data.

[0035] Next, a process using the learned learning model 104a will be described. Here, a process for identifying other behaviors related to a specified user behavior using the Self Attention function of the learning model 104a will be described. Figure 7 is a flowchart showing the process performed by the behavior analysis device 100.

[0036] Processes S201 to S204 are generally the same as processes S101 to S104. That is, in the behavior analysis device 100, the time-series data acquisition unit 101 acquires behavior history data of a user and threshold parameters. The time-series data acquisition unit 101 further acquires a threshold for attention weighting and behavior target information of the user to be compared. The time-series data acquisition unit 101 then acquires behavior identifiers and the like from the behavior history data as time-series data, the generalization processing unit 102 performs generalization processing, and the data processing unit 103 inserts behavior identifiers indicating breaks into positions in the time-series data that satisfy predetermined conditions.

[0037] The BERT processing unit 105 inputs the time-series data including the action identifier and the delimiter identifier into the learning model 104a, and acquires the attention weight for each piece of time-series data (S205).

[0038] The range selection unit 106 selects behavior history data corresponding to the pre-specified user's time series data from all the time series data based on the attention weight for each combination of time series data and the threshold value for the attention weight received in advance (S206). That is, the range selection unit 106 selects time series data (behavior history data) whose attention weight is equal to or greater than the threshold value from among the attention weights between the behavior target information (temporal series data) to be compared and each time series data received in process S201.

[0039] The output unit 107 outputs the selected time-series data (behavior history data) and attention weights (S207).

[0040] In this way, other behavior history data related to the behavior target information can be selected.

[0041] Next, the generalization process will be explained using FIGS. 8 to 11. FIG. 8 is a diagram showing the behavior history database 101a. As shown in the diagram, the behavior history database 101a associates a user ID, a date and time, and categories 1 to 4. The user ID is an identifier for identifying the user. The date and time indicate the date and time when the user performed the behavior. In the diagram, the date and time are shown, but only the date may be shown. Categories 1 to 4 indicate the classification categories of the user's behavior. Category 1 indicates web access or a call to a call center. Category 2 indicates a classification such as a corporate website, an OLT (online procedure), or a general IC (information center). Category 3 indicates a classification such as web browsing or an incoming call (telephone). Category 4 indicates a classification of specific examples of user behavior. In the diagram, it indicates accessing MyPage, accessing a points page, etc.

[0042] FIG. 9 is a diagram showing the aggregation of the occurrence frequency for each category for the behavior history data of all users over a predetermined period. For example, the diagram shows that the frequency of actions taken on Category 1: WEB, Category 2: Corporate Site, Category 3: Browsing, and Category 4: My_Page is 20. This frequency is information aggregated from the behavior history of each user recorded in the behavior history database 101a. In the present disclosure, the generalization processing unit 102 aggregates the frequency for each category when performing generalization processing on time-series data.

[0043] FIG. 10 is a diagram in which the frequency of user behavior classifications is tallied for each classification organized by hierarchy. That is, the frequency of each user behavior is tallied for each major classification, medium classification, and minor classification. In the present disclosure, the classifications are divided into Classifications 1 to 4. Classification 1 indicates the classification of the highest concept of user behavior. Classifications 2 to 4 are defined so that the larger the numerical value, the narrower the concept.

[0044] Figure 10(a) is a diagram showing the aggregated frequency of user actions included in Category 1. Category 1 is a category that includes Categories 2 to 4. As shown in the figure, the frequency of user actions classified into either the user's use of the web or the call center is aggregated.

[0045] FIG. 10(b) is a diagram showing the aggregated frequency of user behaviors included in Category 2. Category 2 is a category that includes Category 3 and Category 4. For example, it shows the aggregated frequency of accessing the corporate website and the OLT.

[0046] FIG. 10(c) is a graph showing the aggregated frequency of user actions included in category 3.

[0047] FIG. 11 is a diagram in which the summary tables shown in FIG. 9 and FIG. 10(a) to FIG. 10(c) are merged. In FIG. 11, the categories are merged and then rearranged in descending order of frequency. In the present disclosure, for categories with the same frequency, categories made up of major categories are ranked higher. Furthermore, in the frequency tallied by major category, items in the medium and minor categories are missing. In this missing part, a character string [UNK] indicating that the information is general information is entered. Note that a character string other than [UNK] may also be used. Furthermore, other symbols may be used, and any character string or symbol indicating that the information is general information may be used.

[0048] Fig. 12 is a diagram showing the top 12 categories by frequency of occurrence extracted from the merged summary table shown in Fig. 11. In this disclosure, this is referred to as a management table. Of course, the number is not limited to 12, and may be any number.

[0049] FIG. 13 is a diagram showing generalized behavior identifiers based on the management table (FIG. 12). The generalization processing unit 102 searches the behavior history database 101a for behavior history records that match each classification described in the management table. Then, the generalization processing unit 102 connects each classification in the matching behavior history records with " / " to generate a behavior identifier. For behavior history records in which the upper classification matches but the lower classification does not match, the generalization processing unit 102 replaces the mismatched lower classification with [UNK]. Then, the generalization processing unit 102 connects each classification with " / " to generate a behavior identifier.

[0050] For example, if record R1 (category 1: WEB, category 2: corporate site, category 3: browsing, category 4: My_Page) described in the management table is also described in the behavior history database 101a, a behavior identifier is generated based on that.

[0051] On the other hand, for example, if categories 1 to 3 of record R2 (category 1: WEB, category 2: OLT, category 3: browsing, category 4: [UNK]) described in the management table are also described in the behavior history database 101a, a behavior identifier is generated using category 4: [UNK] regardless of the content of category 4. In FIG. 13, WEB / OLT / browsing / [UNK] is generated as the behavior identifier for behavior history record R2 (see record R21). In record R21, "price plan option" is registered as category 4, but since the frequency of access to this item is low, it is generalized. Note that, of the behavior identifier WEB / OLT / browsing / [UNK], WEB / OLT / browsing corresponds to the behavior category.

[0052] In this way, when the occurrence frequency of each item is low compared to the overall occurrence frequency, the character string [UNK] is used for generalization. In Figure 13, the behavior identifier of record R31 (category 1: WEB, category 2: OLT, category 3: browsing, category 4: confirming contract details) is WEB / OLT / browsing / [UNK], which is the same as the behavior identifier of record R21. Records R21 and R31 are treated as the same behavior in terms of the user's overall behavior.

[0053] In addition, when calculating the relevance in an attention-based encoder model such as BERT, the two actions are treated as the same. Therefore, the calculation process for calculating attention weights can be reduced.

[0054] FIG. 14 shows a diagram in which behavior identifiers indicating divisions based on the time intervals between the behaviors are inserted into the generalized behavior history data.

[0055] As shown in the figure, records R41 to R43 are inserted between user actions. These records R41 to R43 indicate the action identifiers that indicate the divisions shown in FIG. 4. In FIG. 14, divisions are indicated by inserting "SEP." This makes it easier to understand the adjacent relationships between user actions. That is, when the time interval is short, it is considered that there is a close relationship between adjacent actions. On the other hand, when the time interval is long, it is considered that there is not much correlation between the adjacent actions. In this disclosure, the correlation between these adjacent actions is clarified.

[0056] The insertion of behavior identifiers indicating delimiters shown in FIG. 14 is performed on the behavior history data acquired by the time-series data acquisition unit 101, and then the learning unit 104 performs learning processing using BERT.

[0057] The time-series data acquisition unit 101 acquires, for example, 1000 pieces of behavior history data from all behavior history data as time-series data. Then, the generalization processing unit 102 and the data processing unit 103 perform the above-mentioned generalization processing and the processing of inserting behavior identifiers indicating breaks on the 1000 pieces of time-series data.

[0058] The learning unit 104 performs fill-in-the-blank processing and adjacent prediction processing on the time-series data that has undergone these processes. The fill-in-the-blank processing is performed by randomly masking one or more records of the time-series data. The adjacent prediction processing performs adjacent prediction between records. In this way, the learning model 104a is trained.

[0059] Fig. 15 is a diagram showing the calculation and selection process of attention weights from the user's behavior history database 101a. For convenience, Fig. 15(a) shows the date and time when the behavior occurred and the behavior identifier. As shown in Fig. 4, the record to be the behavior segment target (behavior target information) is specified in advance by the operator.

[0060] 15(b) is a diagram showing the association of self-attention weights calculated based on the self-attention mechanism by the BERT processing unit 105. In the figure, the behavioral identifiers WEB / corporate site / browsing / customer support are specified as the behavioral segment targets. The degree of association with these is expressed as the self-attention weight.

[0061] 15(c) is a diagram showing the range selected by the range selection unit 106 when the threshold value of the self-attention weight is 0.3. Behavior identifiers having a self-attention weight of 0.3 or more are selected. Note that, while in the present disclosure, behavior identifiers having a self-attention weight of 0.3 or more are selected, this is not limiting, and behavior identifiers occurring after the oldest occurring behavior identifier among behavior identifiers having a self-attention weight of 0.3 or more may be selected.

[0062] Next, a description will be given of the effects of behavior analysis apparatus 100 of the present disclosure. In the present disclosure, behavior analysis apparatus 100 functions as an information processing device for generalizing time-series data.

[0063] The behavior analysis device 100 of the present disclosure includes a time-series data acquisition unit 101 that acquires non-significant behavior data from one or more behavioral data indicating user behavior, and a generalization processing unit 102 that generalizes the non-significant behavior data.

[0064] In the present disclosure, non-important behavior data indicates behaviors that occur infrequently in the user's behavior history. Then, the generalization processing unit 102 replaces the non-important behavior data with a predetermined character string (e.g., [UNK]) as a generalization process. The generalization processing unit 102 also performs a generalization process on the non-important behavior data to indicate a behavior category. As described above, if there is a behavior identifier WEB / OLT / browsing / [UNK], WEB / OLT / browsing / indicates the behavior category. Note that non-important behavior data may be determined based not only on the frequency of occurrence but also on other factors. For example, data that is deemed unnecessary for learning may be marked in advance as non-important behavior data. Furthermore, data that has been in use for a predetermined period of time may also be treated as non-important behavior data.

[0065] In the present disclosure, behavior analysis apparatus 100 further includes behavior history database 101a (history storage unit) that stores the history of user behavior data. Time-series data acquisition unit 101 then refers to behavior history database 101a to acquire non-important behavior data.

[0066] The BERT processing unit 105, which functions as a machine learning unit, calculates the attention weight between time series data based on a self-attention function, and acquires one or more other time series data that are highly related to any given time series data based on the attention weight.

[0067] With this configuration, it is possible to calculate highly relevant time series data using the self-attention function of the learning model obtained by performing learning processing on a data format suitable for time series data.

[0068] The BERT processing unit 105 acquires time-series data that occurs after one other time-series data that satisfies a predetermined condition, from among other time-series data whose attention weight is equal to or greater than a predetermined value.

[0069] According to this configuration, other time series data after the occurrence of highly related time series data are treated as related. This may include time series data that is not highly related, but since it is time series data surrounded by highly related time series data, it is not completely unrelated. Such time series data can also be included.

[0070] The block diagrams used to explain the above embodiments show functional blocks. These functional blocks (components) are realized by any combination of hardware and / or software. Furthermore, the method for realizing each functional block is not particularly limited. That is, each functional block may be realized using a single device that is physically or logically coupled, or may be realized using two or more physically or logically separated devices that are connected directly or indirectly (for example, by wire, wirelessly, etc.) and these multiple devices. The functional block may also be realized by combining the single device or multiple devices with software.

[0071] Functions include, but are not limited to, judgment, determination, judgment, calculation, computation, processing, derivation, investigation, search, confirmation, reception, transmission, output, access, resolution, selection, election, establishment, comparison, assumption, expectation, consideration, broadcasting, notifying, communicating, forwarding, configuring, reconfiguring, allocation, mapping, and assignment. For example, a functional block (component) that performs transmission is called a transmitting unit or transmitter. As mentioned above, there are no particular limitations on how these functions are implemented.

[0072] For example, a behavior analysis apparatus 100 according to an embodiment of the present disclosure may function as a computer that performs processing of the behavior analysis method of the present disclosure. Fig. 16 is a diagram illustrating an example of the hardware configuration of the behavior analysis apparatus 100 according to an embodiment of the present disclosure. The behavior analysis apparatus 100 described above may be physically configured as a computer device including a processor 1001, a memory 1002, a storage 1003, a communication device 1004, an input device 1005, an output device 1006, a bus 1007, and the like.

[0073] In the following description, the term "apparatus" can be interpreted as a circuit, a device, a unit, etc. The hardware configuration of behavior analysis apparatus 100 may be configured to include one or more of the devices shown in the figures, or may be configured to exclude some of the devices.

[0074] Each function of the behavioral analysis device 100 is realized by loading specified software (programs) onto hardware such as the processor 1001 and memory 1002, causing the processor 1001 to perform calculations, control communication via the communication device 1004, and control at least one of reading and writing data in the memory 1002 and storage 1003.

[0075] The processor 1001 controls the entire computer by running, for example, an operating system. The processor 1001 may be configured by a central processing unit (CPU) including an interface with peripheral devices, a control device, an arithmetic unit, a register, etc. For example, the above-mentioned generalization processing unit 102, data processing unit 103, etc. may be realized by the processor 1001.

[0076] The processor 1001 also reads programs (program codes), software modules, data, etc. from at least one of the storage 1003 and the communication device 1004 into the memory 1002 and executes various processes in accordance with these. The programs used are those that cause a computer to execute at least some of the operations described in the above-described embodiments. For example, the generalization processing unit 102 may be implemented by a control program stored in the memory 1002 and running on the processor 1001, and similar implementations may be made for other functional blocks. While the above-described various processes have been described as being executed by one processor 1001, they may also be executed simultaneously or sequentially by two or more processors 1001. The processor 1001 may be implemented by one or more chips. The programs may also be transmitted from a network via a telecommunications line.

[0077] The memory 1002 is a computer-readable recording medium and may be configured, for example, by at least one of a read-only memory (ROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), a random access memory (RAM), etc. The memory 1002 may also be called a register, a cache, a main memory (primary storage device), etc. The memory 1002 can store executable programs (program codes), software modules, etc. for implementing a behavior analysis method according to one embodiment of the present disclosure.

[0078] Storage 1003 is a computer-readable recording medium, and may be, for example, at least one of an optical disk such as a CD-ROM (Compact Disc ROM), a hard disk drive, a flexible disk, a magneto-optical disk (e.g., a compact disk, a digital versatile disk, a Blu-ray disc), a smart card, a flash memory (e.g., a card, a stick, a key drive), a floppy disk, a magnetic strip, etc. Storage 1003 may also be referred to as an auxiliary storage device. The above-mentioned storage medium may be, for example, a database, a server, or other appropriate medium including at least one of memory 1002 and storage 1003.

[0079] The communication device 1004 is hardware (transmission / reception device) for communicating between computers via at least one of a wired network and a wireless network, and is also referred to as, for example, a network device, a network controller, a network card, or a communication module. The communication device 1004 may be configured to include a high-frequency switch, a duplexer, a filter, a frequency synthesizer, etc., to realize at least one of frequency division duplex (FDD) and time division duplex (TDD). For example, the above-mentioned output unit 107 may be realized by the communication device 1004.

[0080] The input device 1005 is an input device (for example, a keyboard, a mouse, a microphone, a switch, a button, a sensor, etc.) that receives input from the outside. The output device 1006 is an output device (for example, a display, a speaker, an LED lamp, etc.) that outputs to the outside. The input device 1005 and the output device 1006 may be integrated into one device (for example, a touch panel).

[0081] Furthermore, each device, such as the processor 1001 and the memory 1002, is connected by a bus 1007 for communicating information. The bus 1007 may be configured using a single bus, or may be configured using different buses between each device.

[0082] Behavior analysis apparatus 100 may also be configured to include hardware such as a microprocessor, a digital signal processor (DSP), an application specific integrated circuit (ASIC), a programmable logic device (PLD), or a field programmable gate array (FPGA), and some or all of the functional blocks may be realized by the hardware. For example, processor 1001 may be implemented using at least one of these pieces of hardware.

[0083] The notification of information is not limited to the aspects / embodiments described in the present disclosure and may be performed using other methods. For example, the notification of information may be performed by physical layer signaling (e.g., Downlink Control Information (DCI), Uplink Control Information (UCI)), higher layer signaling (e.g., Radio Resource Control (RRC) signaling, Medium Access Control (MAC) signaling, broadcast information (Master Information Block (MIB), System Information Block (SIB))), other signals, or a combination thereof. Furthermore, the RRC signaling may be referred to as an RRC message, and may be, for example, an RRC Connection Setup message, an RRC Connection Reconfiguration message, or the like.

[0084] The order of the procedures, sequences, flowcharts, etc. of each aspect / embodiment described in this disclosure may be changed unless it is consistent. For example, the methods described in this disclosure present elements of various steps using an example order, and are not limited to the particular order presented.

[0085] Input and output information may be stored in a specific location (for example, memory) or may be managed using a management table. Input and output information may be overwritten, updated, or added to. Output information may be deleted. Input information may be sent to another device.

[0086] The determination may be made based on a value represented by one bit (0 or 1), a Boolean value (true or false), or a numerical comparison (e.g., comparison with a predetermined value).

[0087] Each aspect / embodiment described in this disclosure may be used alone, in combination, or switched depending on the implementation. Furthermore, notification of predetermined information (e.g., notification that "X is true") is not limited to being done explicitly, but may be done implicitly (e.g., by not notifying the predetermined information).

[0088] Although the present disclosure has been described in detail above, it is clear to those skilled in the art that the present disclosure is not limited to the embodiments described herein. The present disclosure can be implemented in modified and altered forms without departing from the spirit and scope of the present disclosure as defined by the claims. Therefore, the description of the present disclosure is intended to be illustrative and does not have any limiting meaning on the present disclosure.

[0089] Software shall be construed broadly to mean instructions, instruction sets, code, code segments, program code, programs, subprograms, software modules, applications, software applications, software packages, routines, subroutines, objects, executable files, threads of execution, procedures, functions, etc., whether referred to as software, firmware, middleware, microcode, hardware description language, or otherwise.

[0090] Software, instructions, information, etc. may also be transmitted or received over a transmission medium. For example, if software is transmitted from a website, server, or other remote source using wired technologies (such as coaxial cable, fiber optic cable, twisted pair, Digital Subscriber Line (DSL)), and / or wireless technologies (such as infrared, microwave), these wired and / or wireless technologies are included within the definition of transmission media.

[0091] The information, signals, etc. described in this disclosure may be represented using any of a variety of different technologies. For example, data, instructions, commands, information, signals, bits, symbols, chips, etc. that may be referred to throughout the above description may be represented by voltages, currents, electromagnetic waves, magnetic fields or magnetic particles, optical fields or photons, or any combination thereof.

[0092] In addition, terms explained in this disclosure and terms necessary for understanding this disclosure may be replaced with terms having the same or similar meanings.

[0093] Furthermore, the information, parameters, etc. described in the present disclosure may be expressed using absolute values, relative values ​​from a predetermined value, or other corresponding information. For example, a radio resource may be indicated by an index.

[0094] As used in this disclosure, the terms "determining" and "determining" may encompass a wide variety of actions. "Determining" and "determining" may include, for example, judging, calculating, computing, processing, deriving, investigating, looking up, searching, inquiring (e.g., searching in a table, database, or other data structure), ascertaining, and the like. "Determining" and "determining" may also include receiving (e.g., receiving information), transmitting (e.g., sending information), input, output, accessing (e.g., accessing data in memory), and the like. Furthermore, "judgment" and "decision" can include regarding resolving, selecting, choosing, establishing, comparing, etc. as having been "judged" or "decided." In other words, "judgment" and "decision" can include regarding some action as having been "judged" or "decided." Furthermore, "judgment (decision)" can be interpreted as "assuming," "expecting," "considering," etc.

[0095] The terms "connected," "coupled," or any variation thereof, refer to any direct or indirect connection or coupling between two or more elements, and may include the presence of one or more intermediate elements between two elements that are "connected" or "coupled" to each other. The coupling or connection between elements may be physical, logical, or a combination thereof. For example, "connected" may be read as "access." As used in this disclosure, two elements may be considered to be "connected" or "coupled" to each other using one or more wires, cables, and / or printed electrical connections, as well as electromagnetic energy having wavelengths in the radio frequency range, microwave range, and optical (both visible and invisible) range, as some non-limiting and non-exhaustive examples.

[0096] As used in this disclosure, the phrase "based on" does not mean "based only on," unless expressly stated otherwise. In other words, the phrase "based on" means both "based only on" and "based at least on."

[0097] When used in this disclosure, the terms "include," "including," and variations thereof are intended to be inclusive, similar to the term "comprising." Furthermore, when used in this disclosure, the term "or" is not intended to be an exclusive or.

[0098] In this disclosure, where articles are added by translation, such as a, an, and the in English, the disclosure may include that the nouns following these articles are in the plural form.

[0099] In the present disclosure, the term "A and B are different" may mean "A and B are different from each other." The term may also mean "A and B are each different from C." Terms such as "separate" and "coupled" may also be interpreted in the same way as "different." [Explanation of symbols]

[0100] 100...Behavioral analysis device, 200...PC, 300...WEB server, 101a...Behavioral history database, 101...Time series data acquisition unit, 102...Generalization processing unit, 103...Data processing unit, 104...Learning unit, 104a...Machine learning model, 105...BERT processing unit, 106...Range selection unit, 107...Output unit.

Claims

1. a machine learning processing unit that performs processing related to machine learning using time-series data including the generalized non-significant behavior data; Equipped with The generalization process is a process of replacing the non-important behavior data with a predetermined character string, The machine learning processing unit is Calculate the weight of attention between the time-series data including the non-important behavior data based on a self-attention function; acquiring one or more other time series data highly related to the given time series data based on the attention weight; Information processing device.

2. The machine learning processing unit is Identifying one other behavioral data for which a predetermined time has elapsed from among the other time-series data determined to be highly relevant based on the attention weight; Obtaining one or more pieces of time series data that occurred after the one other piece of time series data; The information processing device according to claim 1 .

Images