JP2010020731A - Information analyzing device and information analyzing program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide an information analyzing device and an information analyzing program which quickly and surely allow a user to find out information that highly likely leads to the trend in the future. <P>SOLUTION: The information analyzing device 3 is provided with a monitoring part 32b for number of appearance times and a trend detection part 32c. The monitoring part 32b acquires trend-candidate information that a trend-word candidate and a category, to which the trend-word candidate belongs, are correlated with each other while acquiring information from an information providing device 4 via a network 2 so as to count a time-series transition of the number of appearance times of the trend-word candidate included in the acquired information for each category to which a first information provider providing the information belongs. The trend detection part 32c detects that the trend-word candidate is in a specific stage of the trend on the basis of changes in the number of appearance times of the trend-word candidate counted by the monitoring part 32b about a category different from the category corresponding to the trend-word candidate specified on the basis of the trend-candidate information. <P>COPYRIGHT: (C)2010,JPO&INPIT

Description

  The present invention relates to an information analysis apparatus and an information analysis program for analyzing information provided from an information provider such as a blogger who publishes a blog on the WEB.

  Conventionally, by extracting goods, people, or services that are expected to develop into a large-scale epidemic at an early stage of the epidemic, new market development, logistics management based on appropriate demand forecasts, and investment information It is considered that business activities can be promoted more effectively.

  In recent years, the general public has not only obtained information using the Internet, but has also transmitted various information related to work, hobbies, society, etc. via blogs and SNS (Social Network Service). As such, the influence of the Internet on the epidemic is drawing attention. Along with this, a method for discovering trendy items based on information on the Web has been proposed.

  For example, in a service in which a user accesses a predetermined Web server to acquire information, the user group that has quickly accessed information on items that have already been popular is grasped, and the information that the user group is currently accessing A trendy item prediction method has been proposed that predicts as the next trendy item (see, for example, Patent Document 1).

  In addition, a trend prediction device has been proposed that selects keywords related to a trend in the near future based on the degree of attention of keywords extracted from information such as blog articles, the frequency of appearance, and the proportion of positive opinions about the keywords. (For example, refer to Patent Document 2).

JP 2001-350875 A (paragraphs 0043-0046) JP 2006-227965 A (paragraphs 0042-0054)

  In general, in marketing theory, it is said that in order for an item to become a big fashion, it is necessary to attract not only those who are interested in the category to which the item belongs, but also those who are interested in other categories. It has been broken.

  However, according to the method described in Patent Document 1, whether an item newly focused on by a user group that has quickly focused on an item that has been popular in the past remains a small trend that only attracts the attention of the user group, It is not possible to judge whether it has attracted attention beyond the level of a pandemic.

  Further, in the method described in Patent Document 2, who is the keyword provider, who is attracting attention to the keyword, who the keyword appears in, and who is positive about the keyword. It does not consider the attribute of the sender of information about keywords, such as who is negative. Therefore, if the attention level and appearance frequency do not increase over all categories, the possibility of a fashion cannot be detected, and it is impossible to grasp the situation where attention starts to spread from a predetermined category to another category.

  Thus, in the conventional method, the mechanism for discovering a fashion item early and surely was insufficient.

  The present invention has been made in view of the above, and provides an information analysis apparatus and an information analysis program that can quickly and reliably find information that is likely to lead to a trend in the future. Objective.

  According to the information analysis apparatus or the information analysis program according to claim 1 or 5, the natural language in the provided information provided by the first information provider belonging to a category different from the category to which the natural language to be analyzed belongs. When the rate of change in the cumulative number of occurrences exceeds a certain threshold, the detection means determines that the natural language is in the early stage of the epidemic, so it can spread to other fields different from the field to which the natural language belongs. It can be detected quickly. As a result, the natural language can be discovered early and reliably at a stage where attention has been paid in a plurality of fields and before it is widely popularized by being taken up by mass media such as magazines and television.

  Today, not only when positive opinions about items increase, but also when negative opinions (negative campaigns) expand, it is important to detect and prevent this early on to prevent information risk It is important for management. The present information analysis apparatus or information analysis program is also effective in detecting such an expansion of negative opinions early and reliably. That is, when the rate of change in the number of appearances related to the appearance of the natural language to be analyzed exceeds a certain threshold, the detection means determines that the natural language is in the initial stage of the negative campaign. And the expansion of negative campaigns can be detected early and reliably.

  Further, according to the information analysis device of claim 2, when the rate of change of the accumulated number of occurrences of natural language in all provided information provided by the information provider exceeds a certain threshold, the natural language is Since the detection means determines that it is in the early stage of the epidemic, it is possible to reliably find a natural language that is in a stage where it has begun to prevail widely and generally.

  In addition, according to the information analysis apparatus of the third aspect, the rate of change of the natural language appearance number cumulative value in the provided information provided by the first information provider belonging to the same category as the category to which the natural language belongs is obtained. When a certain threshold is exceeded, the detection means determines that the natural language is in the early stage of the epidemic, so early discovery of the natural language that has begun to attract attention by the first information providers who are good at the same field can do.

  In addition, according to the information analysis apparatus of the fourth aspect, depending on whether or not the provided information includes a natural language detected to be in a specific stage of fashion, the information provider who provided the provided information Since the degree of influence in information provision is evaluated, the first information provider is identified while always responding to the latest fashion, and the natural language that is likely to be the next fashion based on the information provided by the first information provider is selected. Can be extracted.

  Embodiments of an information analysis apparatus and an information analysis program according to the present invention will be described below in detail with reference to the accompanying drawings.

(Constitution)
First, the configuration of the information analysis system will be described. FIG. 1 is an explanatory diagram conceptually showing the overall configuration of the information analysis system. The information analysis system 1 includes an information analysis device 3, a plurality of information providing devices 4, a mass media data server 5, a POS (Point of Sale) data server 6, and a terminal device 7 via a network 2 such as the Internet. Are connected so that they can communicate with each other.

(Configuration-Information analyzer)
The configuration of the information analysis device 3 will be described. The information analysis apparatus 3 is configured by connecting a storage unit 30, a network interface (hereinafter “network IF”) 31, and a control unit 32 so that they can communicate with each other via a bus 33.

  The storage unit 30 is configured by, for example, a hard disk or other recording medium, and includes an extraction source table 30a, a transition database (hereinafter “DB”) 30b, a ranking DB 30c, a mass media DB 30d, and a POS data DB 30e.

  The network IF 31 is a communication control interface for the information analysis apparatus 3 to perform communication via the network 2, and is an input unit and an output unit. The network IF 31 is configured as a network board, for example.

  The control unit 32 is a control unit that controls each unit of the information analysis device 3, and functionally conceptually, the trend word candidate extraction unit 32a, the appearance number monitoring unit 32b, the trend detection unit 32c, the transition estimation unit 32d, and the extraction source evaluation A processing unit 32e and an extraction source update unit 32f are provided. Specifically, the control unit 32 includes a CPU (Central Processing Unit), various programs interpreted on the CPU (control programs such as an OS, programs that define various processing procedures, and the like), and And an internal memory for storing required programs and required data.

  Next, specific contents of the information stored in the storage unit 30 described above will be described. The extraction source table 30a is extraction source information storage means for storing information about the extraction source of the natural language extracted from the information providing device 4 by the information analysis device 3, and as illustrated in FIG. “Blogger ID”, “URL”, “special category”, and “class” are provided as items, and information corresponding to these items is stored in association with each other. The information stored corresponding to the item “blogger ID” is identification information for uniquely identifying the information provider from which the natural language is extracted, for example, “BID001” assigned to each information provider. The identification symbol can be used. The information stored corresponding to the item “URL” is information indicating the storage location of the provision information provided from each information provider. For example, URL address, RSS (Rich Site Summary or RDF Site Summary) information, etc. Can be used. The information stored corresponding to the item “special category” is information for specifying the category to which each information provider belongs according to the content of the provided information to be provided. For example, “Gourmet”, “Music”, etc. Category information can be used, and a plurality of category information can be stored in association with one blogger ID. The information stored in correspondence with the item “class” is information for specifying the degree of influence of the information provider in providing information, for example, a cyber blogger that pays attention to items that may lead to epidemic. "Cyber", an opinion leader that has a stronger influence on other information providers than cyber bloggers, although the timing of focusing on items that may become more popular than cyber bloggers is late The information such as “OP” indicating that it is a general blogger not corresponding to an opinion reader or cyber blogger can be used.

  The transition DB 30b of FIG. 1 is trend information for storing trend candidate information in which the natural language to be analyzed and the category to which the natural language belongs are associated with each other, and information regarding the transition of the number of appearances of the natural language in the provided information. As illustrated in FIG. 3, the storage means includes “trend word candidates”, “word categories”, “registration dates”, “counting dates”, “appearance counts”, and “fashion detection flags” as DB items, and information corresponding thereto. Are stored in association with each other. The information stored corresponding to the item “trend word candidate” is the natural language (nouns, verbs, adjectives, phrases) to be analyzed by the information analysis device 3 such as monitoring the transition of the number of occurrences and detecting the specific stage of the fashion. Or a sentence (hereinafter referred to as “trend word candidate”), for example, “Class B gourmet”, “tambourine”, etc. can be stored as shown in FIG. The information stored corresponding to the item “word category” is category information that identifies the category to which each trend word candidate belongs. The information stored in association with the item “registration date” is information specifying the date on which each trend word candidate is stored in association with the item “trend word candidate”. The information stored corresponding to the item “counting date” is information that identifies the date on which the number of occurrences of each trend word candidate in the provided information is counted. The information stored corresponding to the item “number of appearances” is information that specifies the number of occurrences of each trend word candidate in the provided information. For example, for each category (corresponding to the items “gourmet”, “music”, etc. in FIG. 3) to which the first information provider who provided the provision information belongs, from the number of appearances (number of appearances in FIG. 3) and the registration date The cumulative number of occurrences up to each counting date (cumulative in FIG. 3) was evaluated when each trend word candidate was positively evaluated (positive in FIG. 3) and negatively evaluated (negative). Each case (negative in FIG. 3) is stored. Furthermore, the total number of appearances of all categories and the total number of appearances are stored. The information stored in association with the item “fashion detection flag” uniquely identifies whether or not each trend word candidate is detected by the fashion detection unit 32c to be in a positive or negative fashion specification stage. For example, “0” can be stored when not detected, and “1” can be stored when detected. Each information stored in the transition DB 30b is acquired by the control unit 32 from the information providing apparatus 4 via the network 2 in the discovery process, the monitoring process, and the detection process described later, and stored in the transition DB 30b.

  The ranking DB 30c in FIG. 1 is ranking information storage means for storing information related to the ranking of the number of occurrences of trend word candidates. As illustrated in FIG. “Trend word candidates” are provided, and information corresponding to these is stored in association with each other. The information stored corresponding to the item “acquisition date” is information specifying the date on which the ranking of the trend word candidate is acquired. The information stored corresponding to the item “ranking” is information that identifies the rank on each ranking acquisition date. The item “trend word candidate” corresponds to the item “trend word candidate” in FIG. The timing and method for storing these pieces of information in the ranking DB 30c are arbitrary. For example, when the number-of-appearance monitoring unit 32b counts the number of appearances of each trend word candidate in the monitoring process to be described later, the order of appearance is based on the count result. Trend word candidates can be ranked and each information can be stored according to the ranking.

  The mass media DB 30d in FIG. 1 is mass media information storage means for storing information on natural language acquired from the mass media data server 5. As illustrated in FIG. It has word categories, “counting dates”, and “number of appearances”, and information corresponding to these is stored in association with each other. The information stored corresponding to the item “number of appearances” is information for specifying the number of each trend word candidate appearing in mass media such as magazines and televisions. For example, each count is counted for each category to which the mass media belongs. The number of appearances in a day (time series in FIG. 4) and the cumulative number of appearances from the registration date to each counting date (cumulative in FIG. 4) can be stored. The timing and method for storing these pieces of information in the mass media DB 30d are arbitrary. For example, in the monitoring process described later, the information is acquired from the mass media data server 5 via the network 2 by the control unit 32 and stored in the mass media DB 30d. Is done.

  The POS data DB 30e shown in FIG. 1 is a POS data information storage unit for storing information related to natural language acquired from the POS data server 6. As illustrated in FIG. “Category”, “Related Products”, “Record Date”, and “Sales”, and information corresponding to these items are stored in association with each other. The item “trend word candidate” and the item “word category” correspond to the item “trend word candidate” and the item “word category” in FIG. 3. The information stored corresponding to the item “related product” is information for specifying a product associated with each trend word candidate, and as illustrated in FIG. 5, for one trend word candidate. It is also possible to store information for specifying a plurality of related products (for example, “Ramen”, “Gyoza”, etc. for “Class B gourmet”). The information stored corresponding to the item “recording date” is information for specifying the recording date of sales for each related product. The information stored in correspondence with the item “sales” is information for specifying the sales of the products related to each trend word candidate. For example, the sales on each recording date for each related product (in FIG. 5). (Time series) and cumulative sales (accumulated in FIG. 5) from the registration date to each recording date can be stored. The timing and method for storing these pieces of information in the POS data DB 30e are arbitrary. For example, in the monitoring process described later, the information is acquired from the POS data server 6 by the control unit 32 via the network 2 and stored in the POS data DB 30e. The

(Configuration-Information provision device)
The information providing device 4 in FIG. 1 can be configured as a known Web server such as an SNS (Social Network Service) server or a server of various blog sites, and the specific content of the provided information is arbitrary. . The information providing device 4 has a function of storing information provided by an information provider and transmitting the stored information to the information analyzing device 3 in response to a request from the information analyzing device 3 or the terminal device 7.

(Configuration-terminal device)
The terminal device 7 in FIG. 1 is a terminal for an information provider to input and output information with the information analysis device 3 or the information providing device 4.

(Configuration-Mass Media Data Server and POS Data Server)
The mass media data server 5 in FIG. 1 stores information that specifies the number of occurrences of each trend word candidate in a mass media such as a magazine or a television in association with a word category, an appearance date, or the like. The POS data server 6 stores information for specifying the transition of sales of products related to each trend word candidate in association with the word category, sales record date, and the like. These servers have a function of transmitting stored information to the information analysis device 3 or the terminal device 7 in response to a request from the information analysis device 3, and record, for example, mass media appearance log data. Since it can be configured as a known server such as a data server or a Web server, detailed description thereof is omitted.

(Processing content)
Next, the processing performed in the information analysis apparatus 3 shown in FIG. 1 will be described broadly as extraction source update processing, discovery processing, monitoring processing, detection processing, estimation processing, and evaluation processing.

(Processing content-Extraction source update process)
First, extraction source update processing for updating the contents of information stored in the extraction source table 30a will be described. FIG. 6 is a flowchart showing the flow of the extraction source update process. By repeatedly executing this extraction source update processing in parallel with the discovery processing and monitoring processing described later, it is possible to always refer to the extraction source table 30a based on the latest information in the discovery processing and monitoring processing. Become.

  As a premise of this extraction source update process, an administrator or the like specifies a blog of a famous information provider by an arbitrary method, and then sets information on the information provider and the blog in the extraction source table 30a. For example, a blogger ID is given to this information provider to make the information of the item “Blogger ID”, the URL address and RSS of this blog are set as the information of the item “URL”, and the category and class of this blog are specified. Each is set as information on the item “special category” and the item “class”. Hereinafter, the blog initialized in this way is referred to as “initial blog”.

  Under such a premise, when the information providing apparatus is activated, the extraction source update unit 32f refers to the extraction source table 30a and performs crawling with respect to the information providing apparatus 4, thereby corresponding to the item “URL”. Link to the stored URL address or other blog linked to the initial setting blog specified by RSS (hereinafter referred to as “primary link source blog”), or link to this primary link source. Identify other blogs (hereinafter referred to as “secondary link source blogs”). Thereafter, in the same manner, the hierarchy of links to be crawled is further deepened, and another blog (hereinafter referred to as “nth-order link source blog”) that links to the n−1th order link source is specified (step SA-1). ). Hereinafter, n-order link source blogs from these primary link source blogs are collectively referred to as “link source blogs”.

  Subsequently, the extraction source update unit 32f identifies, from among the link source blogs, link source blogs (hereinafter referred to as “access multiple link source blogs”) in which the number of accesses to the link source blog is a predetermined value or more (step SA). -2).

  Then, the extraction source update unit 32f updates the extraction source table 30a by, for example, adding information related to the accessed multiple link source blog to the extraction source table 30a. Specifically, the extraction source update unit 32f issues a blogger ID to the information provider of the access multiple link source blog by a predetermined method, and sets the blogger ID and the URL address and RSS of the access multiple link source blog as items. Information corresponding to “blogger ID” and item “URL” is added (step SA-3). In addition, by setting the information set in the item “good category” of the initial blog as it is in the item “special category” of the accessed many link source blog, the category of the initial setting blog is changed to the Propagate to category (step SA-4).

  Subsequently, the extraction source update unit 32f specifies the class of the accessed many link source blog, and sets it to the item “class” of the extraction source table 30a (step SA-5). For this reason, first, regarding an item that has actually been popular in the past, a natural language (specifically, the name of the item) that identifies the item and a trend date are acquired. Such information can be acquired from, for example, a record in which the trend detection flag “1” is set in the transition DB 30b. Then, the information of the access multiple link source blog acquired by crawling is analyzed in natural language, the day when the natural language for specifying the item first appears in the access multiple link source blog, the specified date, Information to be set in the item “class” is determined to be “cyber”, “OP”, or “general” according to the context with the actual epidemic date acquired from the transition DB 30b. For example, “Cyber” if the specified date is 3 days or more before the actual epidemic date, “OP” if it is 3 days before or after 3 days, and “General” if it is 4 days or later. To do. The blog set in the extraction source table 30a updated in this way is set as an initial setting blog, and thereafter, similarly, steps SA-1 to SA-5 are recursively performed.

(Processing content-discovery processing)
Next, a discovery process for discovering trend word candidates from provided information will be described. FIG. 7 is a flowchart showing the flow of the discovery process. By causing the information analysis device 3 to periodically perform a discovery process, it is possible to always find a trend word candidate having a fashion possibility based on the latest information.

  When the information analysis device 3 is activated, the trend word candidate extraction unit 32a performs crawling on the information providing device 4 (step SB-1). In this case, the trend word candidate extraction unit 32a refers to the extraction source table 30a (step SB-2), and information stored corresponding to the item “class” corresponds to “OP” or “cyber”. Information about the provider (hereinafter referred to as the first information provider), which is the source of trend word candidate extraction, from the provision information stored in the storage location specified by the information stored corresponding to the item “URL” (Step SB-3).

  Subsequently, the trend word candidate extraction unit 32a analyzes the content of the acquired provided information using a known natural language processing method such as morphological analysis or syntax analysis (step SB-4). As a result, when an unknown word is included in the provided information (step SB-5, Yes), the unknown word is stored as a trend word candidate in association with the item “trend word candidate” in the transition DB 30b, and the unknown The word category of the word and the date and time when the unknown word was found are stored in the transition DB 30b in association with the items “word category” and “registration date”, respectively (step SB-6). The unknown word determination method is arbitrary. For example, all the words not stored in the transition DB 30b may be determined as unknown words, or the administrator may determine them. The method for specifying the word category of the unknown word is arbitrary. For example, by referring to the extraction source table 30a, the category to which the first information provider that provides the provision information including the unknown word belongs may be specified as the word category. Alternatively, the trend word candidate extraction unit 32a may refer to an existing keyword classification site and determine the category to which the unknown word belongs.

  When an unknown word is not included in the provided information at step SB-5 (step SB-5, No), or when information about the unknown word is stored in the transition DB 30b at step SB-6, the trend word candidate extraction unit 32a Determines whether or not all of the crawling information providing apparatuses 4 have been visited (step SB-7), and if it is determined that all have been visited (step SB-7, Yes), the process is terminated.

(Processing content-Monitoring processing)
Next, a monitoring process for monitoring the number of trend word candidates in the provided information will be described. FIG. 8 is a flowchart showing the flow of the monitoring process.

  When the information analysis device 3 is activated, the appearance number monitoring unit 32b performs crawling with respect to the information providing device 4 and acquires the provided information (step SC-1).

  Subsequently, the appearance number monitoring unit 32b identifies the trend word candidate with reference to the transition DB 30b, and whether each trend word candidate is evaluated positively (positive) in each provision information acquired in Step SC-1. It is determined whether the evaluation is negative (negative) (step SC-2). For this determination, a language analysis technique such as a known morphological analysis or dependency analysis can be used. For example, a predetermined modifier (“delicious”, “fun”, “ It is determined that the trend word candidate is positively evaluated, and the modifier is a predetermined modifier having a negative meaning (such as “bad” or “dull”). In this case, it is determined that the trend word candidate is negatively evaluated.

  Then, the number of appearances related to the positive evaluation and the number of appearances related to the negative evaluation are counted, and each count result is associated with the counting date as the number of appearances included in all the provided information, and the item “number of appearances” in the transition DB 30b. (Step SC-3).

  In addition, for the number of occurrences related to positive evaluation of trend word candidates included in the provision information provided by the first information provider and the count result of the number of occurrences related to negative evaluation, the provision information was provided. The category to which each first information provider belongs is specified based on the extraction source table 30a, and the count result for each specified category is stored in the item “number of appearances” in the transition DB 30b (step SC-4).

  Further, based on the number of appearances stored in the transition DB 30b in steps SC-3 and SC-4, the cumulative number of appearances related to the positive evaluation of each trend word candidate included in each category and in all provided information, and negative The cumulative number of appearances related to the evaluation is calculated, each “cumulative number of appearances” is stored in the transition DB 30b in association with the “counting date” (step SC-5), and the monitoring process is terminated.

  It should be noted that the transition of the number of appearances of trend word candidates counted by the number-of-appearance monitoring unit 32b in the monitoring process can be graphed and displayed by known display means. FIG. 9 is a graph showing changes in the number of appearances and the cumulative number of appearances counted by the appearance number monitoring unit 32b for one trend word candidate belonging to the word category “gourmet”, with the horizontal axis representing the date and the vertical axis representing the appearance. Indicates the number or cumulative number of occurrences. FIG. 9A is a graph showing the total number of appearances of trend word candidates included in all provided information. The upper graph shows the number of appearances, the lower graph shows the change in the cumulative number of appearances, and the solid line indicates The number of appearances related to positive evaluations and the broken line show the transition of the number of appearances related to negative evaluations. On the other hand, FIG. 9B is a graph showing changes in the number of appearances of each category of trend word candidates included in the provision information provided by the first information provider, and the solid line represents the number of appearances related to positive evaluation, A broken line indicates the transition of the number of appearances related to the negative evaluation, a white bar graph indicates the cumulative number of appearances related to the positive evaluation, and a black bar graph indicates the transition of the cumulative number of appearances related to the positive evaluation.

(Processing content-detection processing)
Next, a detection process for detecting that a trend word candidate is in a specific stage of a trend will be described. FIG. 10 is a flowchart showing the flow of detection processing.

  For example, when a detection process execution command is input or a predetermined time has elapsed and it is determined that the detection process start timing is reached (Yes at step SD-1), the fashion detection unit 32c refers to the transition DB 30b. (Step SD-2) For each trend word candidate, for each of the positive evaluation and the negative evaluation, obtain the transition of the number of appearances for each category and the transition of the total number of appearances in the entire provided information (Step SD-3). . Then, based on the acquired number-of-appearance transition, it is determined whether each trend word candidate is in a specific stage of positive or negative epidemic (step SD-4).

  Specifically, the determination as to whether or not the trend word is in a specific stage is based on whether or not the trend word candidate is in the initial stage of the trend based on at least one or more criteria in the trend word candidate appearance transition. judge. A specific example of this criterion will be shown below.

  The rate of change in a certain period until the execution date of the detection process calculated by the trend detection unit 32c with respect to the cumulative value of the total number of occurrences of trend word candidates stored corresponding to the item “total” in the transition DB 30b is If the set threshold value is exceeded, it can be determined that the attention level of the trend word candidate in society as a whole has begun to rise, so it is determined that the trend word candidate is in the early stage of the epidemic. Can be made. For example, in the example shown in FIG. 9A, the rate of change (slope) of the cumulative number of occurrences related to positive evaluation has increased since March 31, 2008, and the rate of change exceeds a predetermined threshold. When it exceeds, it can be determined that the trend word candidate corresponding to the graph is in the initial stage of positive fashion.

  Further, the rate of change calculated by the trend detection unit 32c with respect to the cumulative value of the number of appearances of the trend word candidate counted for the same category as the word category to which the trend word candidate belongs exceeds a preset threshold. In this case, in the group of first information providers who have high homogeneity in the field of interest, it can be determined that the trend word candidate's attention has begun to rise, so it is in the early stage of the epidemic Can be determined. For example, FIG. 9 shows the transition of the number of trend word candidates belonging to the word category “gourmet”. In FIG. 9B, the provision provided by the first providing information person belonging to the category “gourmet”. The rate of change in the number of cumulative appearances related to positive evaluation in information has already been at a high level since February 9, 2008. That is, since it is considered that the first information provider belonging to the same category as the word category “gourmet” of the trend word candidates is beginning to be positively attracted, it can be determined that the current stage is in the early stage of a positive trend. .

  Further, a threshold value at which the rate of change calculated by the trend detection unit 32c with respect to the cumulative value of the number of occurrences of the trend word candidate counted for any category different from the word category to which the trend word candidate belongs is set in advance. If it exceeds, it is considered that the spread to other fields different from the field to which the trend word candidate belongs has started, so that it can be determined that the trend is in the initial stage. For example, in FIG. 9B, the cumulative number of appearances related to positive evaluation in the provided information provided by the first provided information person belonging to “entertainment” which is a category different from the word category “gourmet” to which the trend word candidate belongs. The rate of change has increased as of March 8, 2008. In other words, attention to the trend word candidates belonging to “gourmet” has spread to other fields, and it is considered that the first information providers who are good at items in the performing arts field are beginning to attract positive attention. be able to. Therefore, it can be determined that the positive epidemic is in the initial stage. In addition, the rate of change of the cumulative number of appearances related to negative evaluation in the provided information provided by the first provided information person belonging to the category “music” is large as of March 8, 2008. Therefore, it can be considered that negative attention has begun to be gathered among first information providers who are good at items in the music field. Therefore, it can be determined that the initial stage of the negative epidemic is also applicable. By using such a determination result, it is possible to detect not only a positive epidemic but also a negative epidemic at an early stage, and take necessary measures before the spread of the negative epidemic.

  In addition, you may make it determine using only one type of determination criteria among the three types of determination criteria illustrated above, or you may make it determine combining arbitrarily.

  Returning to FIG. 10, when the trend detection unit 32c determines that any trend word candidate is in a positive or negative trend specific stage (step SD-4, Yes), the transition DB 30b corresponding to the trend word candidate. In the item “Fashion detection flag”, the flag “1” is stored as information indicating that the positive or negative trend is in a specific stage (step SD-5), and the process is terminated. On the other hand, when none of the trend word candidates is in the trend specific stage (step SD-4, No), the trend detection unit 32c ends the process without updating the content of the transition DB 30b.

(Processing content-guess processing)
Next, the estimation process for estimating the future appearance number transition for each trend word candidate will be described. FIG. 11 is a flowchart showing the flow of the estimation process.

  If the transition estimation unit 32d determines that it is the start timing of the estimation process by performing, for example, a command input for executing the estimation process targeting a specific trend word candidate (step SE-1, Yes), the transition DB 30b , The transition of the accumulated value of the total number of appearances stored corresponding to the item “total” for the trend word candidate to be estimated is acquired over a certain period (step SE-2). Subsequently, the transition estimation unit 32d refers to the cumulative value transition of the total number of appearances for the trend word candidates other than the estimation target, and displays a transition similar to the cumulative value transition of the estimation target trend word candidate acquired in step SE-2. A trend word candidate is searched (step SE-3). As the similarity determination method of the cumulative value transition, a known method for performing similarity search of time series data can be used.

  As a result of the search, when trend word candidates having similar cumulative value transitions are extracted (step SE-4, Yes), the transition estimation unit 32d performs estimation based on the cumulative value transitions of the extracted trend word candidates. An estimated value of the cumulative number of occurrences of trend word candidates is specified (step SE-5).

  On the other hand, if a trend word candidate having a similar cumulative value transition is not extracted as a result of the search (step SE-4, No), the period for acquiring the cumulative value transition is changed (step SE-6). ) Again, the transition of the period after the change is acquired for the trend word candidate to be estimated (step SE-2).

  FIG. 12 is a graph showing the cumulative value transition estimated by the transition estimation unit 32d in the estimation process. In FIG. 12, the cumulative value transition of the trend word candidate to be estimated is represented by a solid line, and the cumulative value transition estimated by the transition estimating unit 32d is represented by a dotted line. As shown in FIG. 12, when a plurality of trend word candidates having similar cumulative value transitions are extracted in step SE-5, the average value of each cumulative value transition may be used as the average guess value, and each cumulative value The maximum value and the minimum value of the transition can be specified as the maximum estimated value and the minimum estimated value.

(Processing content-Evaluation processing)
Finally, an evaluation process for evaluating the degree of influence in providing information by the first information provider will be described. FIG. 13 is a flowchart showing the flow of the evaluation process.

  When the extraction source evaluation processing unit 32e determines that it is the start timing of the evaluation process, for example, by performing a command input for periodic evaluation process execution (step SF-1), the positive epidemic is referred to the transition DB 30b. A trend word candidate whose detection flag is “1” (hereinafter, detection word) is specified, and a registration date of the detection word and a count date when the trend detection flag is “1” are specified (step SF). -2). Next, the extraction source evaluation processing unit 32e acquires provided information provided from the information providing apparatus 4 in a period from the registration date of the detection word to the date when the trend detection flag becomes “1”, and the acquired provision Provided information including a detection word is extracted from the information, and an information provider who provides the provided information is specified (step SF-3).

  Subsequently, the extraction source evaluation processing unit 32e refers to the extraction source table 30a (step SF-4), and information stored in association with the item “class” for the specified information provider corresponds to “general”. It is determined whether or not (step SF-5). As a result, if it falls under “general” (step SF-5, Yes), the information provider is newly recognized as a cyber blogger, and the information corresponding to the item “class” is changed from “general” to “cyber”. The word category to which the detected word belongs is added as information corresponding to the item “special category” (step SF-6). On the other hand, if “general” does not apply (step SF-5, No), the information corresponding to the item “class” is not updated, and the word category to which the detection word belongs for the information corresponding to the item “special category”. Is added only (step SF-7).

  Furthermore, based on the information stored in the extraction source table 30a, the extraction source evaluation processing unit 32e identifies a cyber blogger that is good at the same category as the word category to which the detection word belongs, and from the registration date of the detection word Cyber bloggers that do not include the detection word in the provided information provided during the period up to the date when the trend detection flag becomes “1” are extracted (step SF-8). Regarding the extracted cyber bloggers, it is judged that the ability to quickly focus on items that may become epidemic has declined, and the information corresponding to the item “class” for the bloggers is updated from “cyber” to “general” (Step SF-9). The next and subsequent discovery processes are performed using the blogger class updated in this way.

[Modifications to Embodiment]
Although the embodiments of the present invention have been described above, the specific configuration and means of the present invention can be arbitrarily modified and improved within the scope of the technical idea of each invention described in the claims. Can do. Hereinafter, such a modification will be described.

(About discovery processing)
In the embodiment, the processing performed in the information analysis device 3 has been broadly described as discovery processing, monitoring processing, detection processing, estimation processing, and evaluation processing. Of these, discovery, monitoring, and detection processing are performed as one. It can also be executed simultaneously as a process. That is, while performing crawling with respect to the information providing device 4, the control unit 32 is caused to execute in parallel a discovery of a new unknown word, a count of the number of appearances of trend word candidates, or a trend word candidate's trend stage specification. be able to.

(About detection processing)
In the embodiment, the epidemic detection unit 32c is described as detecting that it is in a specific stage of a positive or negative epidemic based on the transition of the number of appearances of the provided information for each of the positive evaluation and the negative evaluation for the trend word candidate. However, the detection may be performed based on other criteria. For example, based on the transition of the total number of positive evaluations and negative evaluations, it is detected that there is a specific stage of the epidemic, and the cumulative number of occurrences of positive evaluation and negative evaluation at the time of detection is compared. Then, it may be determined whether the detected fashion is positive or negative. Alternatively, detection is performed based on the transition of the total number of occurrences of positive and negative evaluations, and the detected epidemic is either positive or negative based on the ratio between the number of positive evaluations and the number of negative evaluations at the time of detection. You may be able to judge whether it is a fashion.

(Modification-About distribution and integration)
Further, each of the electrical components described above is functionally conceptual and does not necessarily need to be physically configured as illustrated. In other words, the specific form of distribution / integration of each part is not limited to the one shown in the figure, and all or a part thereof may be functionally or physically distributed / integrated in arbitrary units according to various loads and usage conditions. Can be configured. For example, the information analysis device 3 and the information providing device 4 may be integrated with each other.

It is explanatory drawing which shows the whole structure of an information analysis system functionally conceptually. 2A is a table showing information stored in the extraction source table, and FIG. 2B is a table showing information stored in the ranking DB. It is a table | surface which shows the information stored in transition DB. It is a table | surface which shows the information stored in mass media DB. It is a table | surface which shows the information stored in POS data DB. It is the flowchart which showed the flow of the extraction source update process. It is the flowchart which showed the flow of the discovery process. It is the flowchart which showed the flow of the monitoring process. FIG. 9A is a graph showing changes in the number of appearances of trend word candidates and the cumulative number of appearances, and FIG. 9A is a graph showing changes in the total number of appearances of trend word candidates included in all provided information. ) Is a graph showing changes in the number of appearances for each category of trend word candidates included in the provision information provided by the first information provider. It is the flowchart which showed the flow of the detection process. It is the flowchart which showed the flow of the estimation process. It is the graph which showed the cumulative value transition estimated by the transition estimation part in the estimation process. It is the flowchart which showed the flow of the evaluation process.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 Information analysis system 2 Network 3 Information analysis apparatus 4 Information provision apparatus 5 Mass media data server 6 POS data server 7 Terminal apparatus 30 Memory | storage part 30a Extraction source table 30b Transition DB
30c ranking DB
30d Mass Media DB
30e POS data DB
31 Network IF
32 Control Unit 32a Trend Word Candidate Extraction Unit 32b Appearance Number Monitoring Unit 32c Epidemic Detection Unit 32d Transition Estimation Unit 32e Extraction Source Evaluation Processing Unit 32f Extraction Source Update Unit 33 Bus

Claims (5)

An information analysis apparatus connected to an information providing apparatus for storing provided information provided by a first information provider so as to be communicable via a network,
Obtain trend candidate information in which a natural language to be analyzed and a category to which the natural language belongs are associated with each other, obtain the provision information from the information provision apparatus via the network, and include in the obtained provision information Counting means for counting the time-series transition of the number of occurrences of the natural language for each category to which the first information provider that provided the provision information belongs,
Based on a change in the number of occurrences of the natural language counted by the counting means for a category different from the category corresponding to the natural language identified based on the trend candidate information, the natural language is in a trend identification stage. Detection means for detecting
An information analysis apparatus comprising: The information providing device stores provision information provided by the first information provider and information providers including the first information provider;
The counting means counts a time-series transition of the total number of appearances of the natural language included in all provided information provided by the information provider,
The detecting means detects that the natural language is in a specific stage of fashion based on a change in the total number of appearances counted by the counting means;
The information analysis apparatus according to claim 1. The detection means detects that the natural language is in a trend identification stage based on a change in the number of occurrences of the natural language counted for the category corresponding to the natural language specified based on the trend candidate information. To do,
The information analysis apparatus according to claim 1 or 2, characterized by the above. In the case where the detection means detects that the natural language is in a specific stage of fashion,
The presence or absence of the natural language in the provided information provided between the start of counting of the natural language by the counting unit and the detection by the detecting unit is determined for each first information provider who provides the provided information. , An evaluation means for evaluating the degree of influence in the information provision of the first information provider based on the determination result,
Based on the degree of influence of the first information provider evaluated by the evaluation means, the first information provider that is the acquisition target of the trend candidate information is selected and provided by the selected first information provider. Trend candidate extraction means for acquiring the provided information from the information providing apparatus via the network and acquiring the trend candidate information based on the acquired provided information;
The information analysis apparatus according to any one of claims 1 to 3, further comprising: An information analysis program for operating an information analysis apparatus as a computer communicably connected via a network with respect to an information provision apparatus storing provision information provided by a first information provider,
Obtain trend candidate information in which a natural language to be analyzed and a category to which the natural language belongs are associated with each other, obtain the provision information from the information provision apparatus via the network, and include in the obtained provision information Counting a time-series transition of the number of appearances of the natural language for each category to which the first information provider that provided the provision information belongs,
Based on a change in the number of occurrences of the natural language counted by the counting means for a category different from the category corresponding to the natural language identified based on the trend candidate information, the natural language is in a trend identification stage. A detection step for detecting
Is executed by the information analysis apparatus.

Images