JP5952241B2 - Information grant apparatus, information grant method, and information grant program - Google Patents

Description

  The present invention relates to an information assigning device that assigns other information to information, and more particularly, to a technique for assigning metadata about contents to a program such as a television or VOD (video on demand).

  Metadata for program contents such as TV and VOD has been provided for a long time. Using information such as voice, subtitles, closed captions, etc., and using technologies such as character recognition, image recognition, voice recognition, etc., identify people, things and events that appear in the video, and recognize them along a time series. Content has been described, indexing for searching and video content classification have been performed. (Refer nonpatent literature 1).

  Thereafter, attempts have been made to add metadata for program content based on remark information input to program content by program viewers on social media.

  For example, according to Non-Patent Document 2, in a scene in which a program viewer is interested, a climax is extracted using two parameters that the number of utterances and the number of utterance character strings in a chat increases. We are proposing a method to extract keywords from the comments contained in the video and use it as metadata for program content.

  Non-Patent Document 3 describes a method for indexing a scene by determining the degree of excitement / disappointment from the number of entries of written text and the appearance frequency of ASCII art in the written text.

  Further, according to Patent Document 1, more appropriate metadata is given by defining and using the importance level of the user based on not only the number of times the user utters the program content but also the number of citations. Yes.

  In the above existing technology for giving metadata, chat information related to program content is used. However, for example, a method using a search log of a search portal site is conceivable. The search log contains a large number of topic words related to “people, things, and events”, and the user tends to search with a search engine when they see or hear such words. For example, by using this tendency, as shown in Patent Document 2, a method of recommending a trendy search query according to the user's interest and supporting the user's information search has been proposed. As described above, it is considered to be very effective to use the search log in order to analyze the topic word.

  In addition, the technique relevant to this invention is disclosed by patent document 3, nonpatent literature 4,5,6, for example.

JP 2012-173774 A JP2012-59182A JP 2004-46775 A

Nishio, 5 others, “Iwanami Lecture Multimedia Informatics 8 Information Structuring and Retrieval”, Iwanami Shoten, 2000, 3.1 Multimedia Environment and Media Analysis Daikuro, 4 others, “Implementation and Evaluation of Automatic Program Metadata Generation System Using Internet Chat”, 2005-AVM-18, Information Processing Society of Japan Research Report, 2005 Miyamori, two others, “Generating Broadcast Program Views Using Viewer Perspectives Based on Program Live Chat”, DEWS 2005 4B-i9, 2005 R. B. Cleveland, W.M. S. Cleveland, J.A. E. McRae, and I.M. Terpenning, “STL: A seaside-trend decomposition procedure based on losses,” Journal of Official Statistics, Vol. 6). No. 1.1990. pp. 3-73. H. Liu, J .; He, Y. Gu, H .; Xiong, and X. Du, “Detecting and Tracking Topics and Events from Web Search Logs”, ACM Transactions on Information Systems (TOIS), Vol. 30, no. 4, November 2012, pp. 1-29. G. Salton and M.M. McGill, “Introduction to Modern Information Retrieval”, McGraw-Hill, 1983.

  However, while the chat information used by the existing technology for giving metadata to the program content is originally obtained from the chat conducted only for the program content, the search log is not only the program content, It contains words searched based on various factors.

  In general, the number of words searched by a search engine often includes periodic fluctuations. For example, words related to leisure facilities tend to be searched more often on weekends, which are holidays, than on weekdays when there are many jobs. Therefore, even if leisure facilities are introduced in the program content and the number of related word searches increases, if the periodic fluctuation is larger, the fluctuation in the number of searches caused by the program content cannot be detected correctly, and the appropriate topic It becomes difficult to extract words. In order to add metadata to program content using a search log, it is important to remove as much as possible the variation in the number of searches caused by other than program content.

  In addition, existing technologies using chat information and search logs only consider the increase in the number of chat utterances and searches. However, it is considered that the more topical the content, not only the number of utterances and the number of searches increases more rapidly, but also decreases more gradually. This is because the topical content is spread over time by word of mouth or the like, and the number of searches is expected to be sustained.

  The present invention solves the above-described problems, and an object thereof is to provide an information providing apparatus, method, and program capable of extracting words related to program broadcasting with higher accuracy and adding appropriate metadata. There is.

An information providing apparatus according to the present invention for solving the above-described problem is an information providing apparatus for providing information related to a program, and includes metadata including text data relating to program contents and date / time data representing a broadcast time of the program from the program data. A metadata extracting means for extracting the text data, a morphological analysis of the text data extracted by the metadata extracting means, a word extracting means for extracting word data from the analysis result, and a word data extracted by the word extracting means A word generation time calculation unit that calculates a word generation time generated in time-series text data obtained from a search log or Twitter, and a word generation time for each word in the word data extracted by the word extraction unit. By quantizing the word occurrence time information calculated by the calculation means, a time series data on the word occurrence frequency is obtained. Create a data h _wi (t), to separate the trend component and the periodic variation component from the created time series data h _wi (t), and extracted irregular fluctuation component included variation due to the program Time-series data h ′ _wi (t) is created, and using the created time-series data h ′ _wi (t) and date / time data extracted by the metadata extraction means, word feature based on time-series data And calculating a word feature based on the document group with reference to a document database storing the document group, and finally using the word feature based on the time series data and the word feature based on the document group. Using the word feature degree calculating means for calculating the word feature degree and the final word feature degree calculated by the word feature degree calculating means, the feature word is extracted from the word data extracted by the word extracting means into the program. Relation And a feature word extracting means for extracting as information to be performed.

Further, the word feature degree calculating means calculates the word feature degree based on the time series data using the time series data h ′ _wi (t) and the date / time data when calculating the word feature degree based on the time series data. Calculate the word feature based on the variation of the time series data including the kurtosis and the skewness, and from the word feature based on the calculated variation of the time series data and the word feature considering the number of occurrence of the word itself It is characterized by calculating word feature based on time series data.

(1) According to the first to fifth aspects of the present invention, the trend fluctuation component and the period fluctuation component, which are causes of fluctuations other than the program broadcast in the time series text data obtained from the search log or Twitter, are separated and removed. Then, irregular fluctuation components that are fluctuation components caused by the program are extracted, and the word feature degree calculated based on the irregular fluctuation components is used, so that words related to the program broadcast can be extracted with higher accuracy and appropriate. Metadata can be added.
(2) According to the inventions as set forth in claims 2, 4, and 5, the word feature degree is based on the kurtosis and the skewness calculated from the distribution of the time series data relating to the word occurrence frequency extracted as the irregular fluctuation component. Therefore, appropriate metadata considering topicality can be assigned.

The block diagram which shows one Example of the information provision apparatus of this invention. The flowchart which shows one embodiment of the information provision method of this invention. Explanatory drawing which shows the example of data extracted by the metadata extraction part of one embodiment of this invention. Explanatory drawing which shows the example of a calculation result of the word generation time calculation part of one embodiment of this invention. The flowchart of the process which the word feature degree calculation part of one embodiment of this invention performs. Explanatory drawing which shows the mode of the process which the word feature degree calculation part of one example of this invention decomposes | disassembles three components from time series data. Explanatory drawing which shows a mode that the word feature degree calculation part of one embodiment of this invention extracted the irregular fluctuation component from raw time series data. An example of extracted word groups, (a) is an explanatory diagram of word groups extracted using raw time series data according to the prior art, (b) is extracted using irregular variation components according to this embodiment example FIG. In one embodiment of the present invention, an explanatory diagram showing a state in which a topic is more important as a decrease after a sudden increase in the number of words is generated, and a large value is assigned to the word feature.

  Hereinafter, embodiments of the present invention will be described with reference to the drawings, but the present invention is not limited to the following embodiments. In the present invention, using time-series text data such as a search log, the program content is provided with appropriate metadata in consideration of topicality, analyzing the time-series text data obtained from the search log, By reducing the causes of fluctuations other than program content, the degree of fluctuation due to more appropriate program content is calculated. Furthermore, not only the increase in the number of searches but also the degree of decrease is taken into account to extract more topical phrases. The present invention can be applied not only to search logs but also to information sources such as Twitter (registered trademark) and time series text data that can refer to time stamps.

  FIG. 1 is a block diagram showing the configuration of an information providing apparatus according to an embodiment of the present invention, and FIG. 2 is a flowchart showing the overall flow of processing performed by the apparatus of FIG.

  In FIG. 1, the information providing apparatus of the present embodiment includes a metadata extraction unit 10 as a metadata extraction unit, a morpheme analysis unit 20 and a word extraction unit 30 as a word extraction unit, and a word generation as a word occurrence time calculation unit. It includes a time calculator 40, a word feature calculator 50 as a word feature calculator, a feature word extractor 60 as a feature word extractor, a time series text DB (database) 70, and a document DB (database) 80. .

  The information providing apparatus in FIG. 1 is configured by a computer, for example, and includes hardware resources of a normal computer, such as a ROM, a RAM, a CPU, an input device, an output device, a communication interface, a hard disk, a recording medium, and a driving device thereof. .

  As a result of the cooperation between the hardware resource and the software resource (OS, application, etc.), as shown in FIG. 1, the information adding device is a metadata extraction unit 10, a morpheme analysis unit 20, a word extraction unit 30, and a word generation A time calculation unit 40, a word feature degree calculation unit 50, a feature word extraction unit 60, a time series text DB 70, and a document DB 80 are implemented.

  It is assumed that the time series text DB 70 and the document DB 80 are constructed in storage means / storage means such as a hard disk or RAM.

  Next, the operation of the apparatus configured as described above will be specifically described. First, the metadata extraction unit 10 acquires program data from the Web, for example, and extracts metadata (step S210). The metadata extracted here represents text data related to the program content, date / time data representing the broadcast time of the program, and the like. For example, as shown in FIG. 3, there are sentences relating to the introduction of a zoo as text data, and date / time information corresponding to each sentence as date / time data. For text data, a closed caption, an electronic program guide (EPG), a telop character recognition result, a speech recognition technology result, and the like are used.

  Next, the morpheme analyzer 20 decomposes the text data in the metadata extracted by the metadata extractor 10 into the smallest word that makes grammatical meaning by analyzing the morpheme, and the result is the word extractor 30. (Step S220). The word extraction unit 30 extracts word data from the morpheme analysis result (step S230). As an extraction method, for example, an arbitrary part of speech such as a noun is extracted based on a morphological analysis result, or a specific expression extraction method proposed in Patent Document 3 or the like is used.

  Subsequently, the word generation time calculation unit 40 calculates the word generation time related to the extracted word data with reference to the time series text DB 70 (step S240). The time series text DB 70 stores a search log, Twitter's Tweet (registered trademark) data, and the like. For example, as shown in FIG. 4, the word occurrence time is calculated as the date and time when each word in the word data occurred in the search log or Tweet. The word occurrence time can be stored once in the storage device and added when the time-series text DB 70 is updated.

  The calculated word occurrence time is transferred to the word feature degree calculation unit 50. Thereafter, the word feature degree calculation unit 50 calculates the word feature degree using the word data, the date / time data, the word occurrence time, and the document DB 80, and passes the result to the feature word extraction unit 60 (step S250). . The feature word extraction unit 60 extracts feature words according to the word feature degree calculated by the word feature degree calculation unit 50 (step S260).

  Details of the process (step S250) executed by the word feature degree calculation unit 50 will be described with reference to the flowchart of FIG.

  The definition of each parameter handled by the word feature degree calculation unit 50 is as follows.

The time series data regarding the number of word occurrences (word occurrence frequency) is _defined as h _wi (t).

The time series data relating to the number of occurrences of words (word occurrence frequency) extracted as irregular fluctuation components is assumed to be h ′ _wi (t).

  The definition of the word feature is as follows.

  In the following explanation, when expressing each of the word feature degrees, it is expressed using the formulas (1) to (5).

First, with respect to the word wi∈W of the index i in the word data W, the time series data relating to the number of word occurrences is calculated by quantizing the word occurrence time information acquired from the word occurrence time calculation unit 40 (step S510, S520). Quantization is calculated as the number of occurrences of the word wi before and after the time t, and the time series data regarding the number of occurrences of the word after the calculation is represented as h _wi (t).

Next, the time-series data h _wi (t) is decomposed into three components of a trend variation component, a periodic variation component, and an irregular variation component as shown in FIG. 6, and the irregular variation component in FIG. 6 (d) is extracted. Time-series data h ′ _wi (t) is created (step S530).

  Since irregular fluctuation components include fluctuations resulting from program broadcasts, it is possible to extract words related to program broadcasts with higher accuracy by detecting a sudden increase in numerical values from irregular fluctuations. The reason will be described next. In general, many time-series data have a factor of a trend fluctuation indicating a long-term fluctuation, a periodic fluctuation indicating a periodic fluctuation of an interval such as a week, a month, or a quarter, and an irregular fluctuation caused by an accidental event. Many of the time series data such as the number of searches handled this time also have these factors. For example, words related to leisure facilities tend to be searched more often on weekends, which are holidays, than on weekdays when there are many jobs. That is, the variation in the number of searches due to this tendency is classified as a periodic variation because it has a certain periodicity at weekly intervals.

On the other hand, when a leisure facility is introduced in a program broadcast, it is predicted that a sudden change in the number of searches will occur in addition to the above-described periodic fluctuation. Therefore, although the titles of programs that are broadcast periodically are exceptions, fluctuations in the number of searches that are basically caused by new contents provided by program broadcasting are included in irregular fluctuations. Therefore, in the present invention, irregular fluctuation components are extracted from the time series data h _wi (t), and the irregular fluctuation components are used for the calculation of the word feature.

  Specifically, the effect obtained by using the irregular fluctuation component will be described with reference to the examples of FIGS. FIG. 7 shows raw time-series data (FIG. 7 (a)) and raw time-series data regarding the occurrence frequency of the words “penguin”, “hot spring”, and “Asahiyama Zoo” in certain time-series text data. The irregular fluctuation component (FIG. 7 (b)) is shown.

  Here, the raw time-series data for each word has the following properties.

  ・ "Penguin" does not include periodic fluctuations, but includes fluctuations due to program broadcasts.

  ・ “Onsen” includes periodic fluctuations and does not include fluctuations due to program broadcasts.

  ・ “Asahiyama Zoo” includes periodic fluctuations and fluctuations due to program broadcasts.

  When feature words are extracted based on a rapid increase in numerical values in these raw time series data, a result as shown in FIG. 8A is obtained. Looking at this result, it is affected by periodic fluctuations when detecting a sudden increase in numerical values, so “hot springs” that are not actually affected by program broadcasts are extracted to the top, or the effects of program broadcasts are affected. It can be seen that the “Penguin” most received has been extracted in the lower order. On the other hand, by using an irregular fluctuation component obtained by removing the cyclic fluctuation component from the raw time series data, compared with the case where the raw data is used as it is, the program broadcasting is affected as shown in FIG. Can be extracted with higher accuracy. The conventional invention for adding metadata to a program uses raw time-series data as it is. On the other hand, in the present invention, it is assumed that irregular fluctuation components extracted from raw time-series data include fluctuations due to program broadcasting, and this is the first attempt to use them for feature word extraction.

As an irregular fluctuation component extraction method, for example, STL (see Non-Patent Document 4) that decomposes time-series data into each element can be used. By using the STL, the time series data h _wi (t) is defined as the sum of each component and is decomposed as in the following equation (6).

h _wi (t) = trend _wi (t) + seasonal _wi (t) + remainder _wi (t) (6)
Here, trend _wi (t) indicates trend variation, seasonal _wi (t) indicates periodic variation, and remainder _wi (t) indicates irregular variation. In the present invention, irregular fluctuation remainder _wi (t) is used as fluctuation caused by program broadcasting. At this time, since remainder _wi (t) includes a negative value, h ′ _wi (t) in the following equation (7) is used as an irregular fluctuation component.

  In addition, in the present invention, a word that has been a hot topic for a long time is extracted by taking into account the degree of decrease after the rapid rise. In general, in a time series data, a state in which a numerical value increases significantly at a certain time is called a burst, and many techniques for detecting a burst have been proposed. In Non-Patent Document 5, time series data is created for the number of clicks on a page URL on the Web, bursts are detected by using entropy obtained from the time series data, and important events are extracted. However, the entropy technique can detect a state with a transient change as a burst, but cannot detect a state whose numerical value persists after a large change. For example, the former is treated as an important event when the number of searches related to a product introduced in a program suddenly drops after a sharp rise, and when the number of searches for a product that has suddenly risen and then gradually decreases.

  In the present invention, as shown in FIG. 9 where the vertical axis represents the number of occurrences of the word and the horizontal axis represents the time, as the decrease after the sudden increase in the number of occurrences of the word is more gradual, the topic is more widely spread and important Assuming that the topic is a topic, a larger value is assigned to the word feature (formula (1)).

In the present invention, the kurtosis and distortion in the distribution of the time series data are calculated based on the following assumptions (a) and (b) in order to calculate the word feature degree (formula (1)) based on the fluctuation of the time series data. Utilize degrees.
(A) The greater the kurtosis, the greater the degree of kurtosis and the degree of gathering of the distribution. Therefore, suddenly large fluctuations occur in the number of word occurrences, which are greatly influenced by program broadcasting.
(B) The greater the skewness, the longer the distribution extends in the positive direction of the time axis, for example, as shown in FIG. 9 (c). It ’s been a long time.

Specifically, the word feature degree (formula (1)) based on the fluctuation of the time series data is used as the time series data h ′ _wi (t) extracted as irregular fluctuation components and the broadcast date and time di regarding the word wi. Calculate as follows.

  The above-described feature level (formula (1)) is determined based on the temporal variation of the number of word occurrences, and does not depend on the number itself. Therefore, even if the number of generated words is small, the word feature degree becomes large depending on the distribution of the time series data. In Non-Patent Document 5, in addition to the above-described method based on entropy, this is solved by introducing a Concern Rate that determines the score according to the number of clicks on the page URL. The present invention also introduces the above-mentioned Concate Rate.

  Here, α is an arbitrary constant for determining the specific gravity of each feature. The expression (2) represents the degree of the number of word searches and the like, and the value is determined based on the number of searches other than wi and wi in the word data W. The above equation (2) is 1 when the occurrence of wi is the most, and 0 when the occurrence of the wi is the least.

  In the present invention, like the Concaten Rate disclosed in Non-Patent Document 5, a score relating to the frequency of time series data (formula (2)) is used. On the other hand, in this embodiment, in addition to that, as a score related to the distribution of time series data, a new score using kurtosis and skewness, that is, a word feature (formula (1)) based on fluctuation of time series data is defined. Introduced word feature degree. This realizes detection that takes into account burst duration, which is difficult with existing methods.

  The word feature (formula (1)) based on the time-series data can detect a highly topical word by considering the influence of program broadcasting on people.

  Note that the calculations of the equations (8) to (14) are executed in step S540 in FIG.

  Further, in the present invention, in order to extract program-specific characteristic words compared with other programs, a word characteristic (expression (4)) based on a document set is calculated (step S550) and used. This is effective in distinguishing words that do not appear as significant fluctuations or word occurrences in time-series data from other unimportant general words. For example, the conventional technique tf-idf (see Non-Patent Document 6) or the technique disclosed in Patent Document 1 can be used for the word feature (formula (4)) based on the document set. The document corpus stored in the document DB 80 to be referred to for calculating the word feature (formula (4)) stores, for example, metadata of programs broadcast so far, and the metadata of each program is stored in one document. There is a method to use as.

Finally, after calculating the word feature based on the time series data (formula (3)) and the word feature based on the document corpus (formula (4)), the final word feature shown in the following formula (15) f _wi is calculated.

Here, β is an arbitrary constant for determining the specific gravity of each feature. f _wi is calculated for all words wiεW (steps S560 to S580).

The feature word extraction unit 60 extracts feature words using the final word feature degree f _wi calculated by the word feature degree calculation unit 50 by the above processing (step S260 in FIG. 2). For example, as shown in FIG. 8, the feature word extraction unit 60 creates and outputs a list in which words are sorted in descending order of the word feature degree f _wi .

  As described above, in this embodiment, metadata is added to program content by using time series text data such as search logs and Twitter including various fluctuation factors as well as chat related to program broadcasting so far. Decompose time series data calculated from logs and Twitter, use irregular fluctuations as fluctuations caused by program broadcasts, and consider the degree of decrease in numbers after a sudden rise in the calculation of the degree of fluctuation in time series data Configured.

  Thereby, metadata considering the topicality can be given to the program content. Since search engines often search for words related to “persons, things, and events” of interest, the search log is very useful for extracting words that are topics. However, fluctuations in the time series data obtained from the search log are caused not only by the program broadcast but also by various events. Many other time-series text data such as Twitter have similar properties. According to this embodiment, it is possible to calculate a more appropriate degree of fluctuation by separating time-series data obtained from a search log and reducing the cause of fluctuations other than program broadcasting. In addition to the increase in the number of generated words, a more topical phrase can be extracted by considering the degree of decrease.

  In addition, a part or all of the functions of each unit in the information providing apparatus of the present embodiment can be configured by a computer program, and the program can be executed using the computer to realize the present invention. It goes without saying that the procedure in the information assigning method can be constituted by a computer program, and the program can be executed by the computer, and the computer-readable recording medium, for example, FD, can be realized by the computer. (Floppy (registered trademark) Disk), MO (Magneto-Optical disk), ROM (Read Only Memory), memory card, CD (Compact Disk) -ROM, DVD (Digital Versatile D) sk) -ROM, CD-R, CD-RW, HDD, and recorded in a removable disk, or stored, it is possible or distribute. It is also possible to provide the above program through a network such as the Internet or electronic mail.

DESCRIPTION OF SYMBOLS 10 ... Metadata extraction part 20 ... Morphological analysis part 30 ... Word extraction part 40 ... Word generation time calculation part 50 ... Word feature degree calculation part 60 ... Feature word extraction part 70 ... Time series text DB
80 ... Document DB

Claims (5)

An information granting device for granting information related to a program,
Metadata extraction means for extracting, from program data, metadata including text data relating to program contents and date / time data representing the broadcast time of the program;
Morphological analysis of the text data extracted by the metadata extraction means, and word extraction means for extracting word data from the analysis result;
A word occurrence time calculating means for calculating a word occurrence time for each word of the word data extracted by the word extracting means, which is generated in time-series text data obtained from a search log or Twitter;
Time-series data h _wi (t) relating to the word occurrence frequency is created by quantizing the word occurrence time information calculated by the word occurrence time calculation means for the words of the word data extracted by the word extraction means. and separates the trend component and the periodic variation component from the created time series data h _wi (t), when extracting the random variation component that contains the variation due to the program series data _h'wi (t ), And using the created time series data h ′ _wi (t) and the date and time data extracted by the metadata extraction means, the word feature degree based on the time series data is calculated, and the document group is stored. The word feature based on the document group is calculated with reference to the document database, and the final word feature is calculated using the word feature based on the time series data and the word feature based on the document group. A word features calculation means for calculating,
Feature word extraction means for extracting a feature word from the word data extracted by the word extraction means as information related to a program using the final word feature degree calculated by the word feature degree calculation means;
An information providing apparatus comprising: The word feature degree calculating means includes:
When calculating word features based on time-series data,
Using the time-series data h ′ _wi (t) and the date / time data, calculate a word feature based on fluctuations in the time-series data, including kurtosis and skewness calculated from the distribution of the time-series data,
The word feature based on the time series data is calculated from the word feature based on the calculated variation of the time series data and the word feature considering the number of occurrences of words itself. Information grant device. An information providing method for providing information related to a program,
A step of extracting metadata including text data relating to program contents and date / time data representing broadcast time of the program from the program data;
A word extracting unit performs morphological analysis on the text data extracted by the metadata extracting unit, and extracts word data from the analysis result;
A word occurrence time calculating means for calculating a word occurrence time in which each word of the word data extracted by the word extracting means is generated in time-series text data obtained from a search log or Twitter;
The word feature degree calculation means quantizes the word occurrence time information calculated by the word occurrence time calculation means for the words of the word data extracted by the word extraction means, so that time series data relating to the word occurrence frequency is obtained. creating h _wi (t);
Time-series data in which the word feature degree calculation means extracts the irregular variation component including the variation caused by the program by separating the trend variation component and the periodic variation component from the created time-series data h _wi (t) creating h ′ _wi (t);
A word feature degree calculating means calculating a word feature degree based on the time series data using the created time series data h ′ _wi (t) and the date and time data extracted by the metadata extracting means;
A word feature degree calculating means for calculating a word feature degree based on a document group with reference to a document database storing the document group;
A word feature calculation means calculating a final word feature using the word feature based on the time-series data and the word feature based on a document group;
A step of extracting a feature word from the word data extracted by the word extraction unit as information related to a program, using the final word feature calculated by the word feature calculation unit; When,
An information providing method characterized by comprising: The step of calculating the word feature based on the time series data by the word feature calculation means,
Using the time-series data h ′ _wi (t) and the date / time data, calculate a word feature based on fluctuations in the time-series data, including kurtosis and skewness calculated from the distribution of the time-series data,
The word feature based on the time series data is calculated from the word feature based on the calculated variation of the time series data and the word feature considering the number of occurrences of words itself. Information grant method.   The information provision program which makes a computer function as each means of Claim 1 or 2.

Images