JP6554306B2 - Information processing system, information processing method, and computer program - Google Patents

Description

  The present invention relates to an information processing system, an information processing method, and a computer program.

  2. Description of the Related Art Conventionally, an information search apparatus that accepts a keyword input from a user and realizes information presentation considering diversity without browsing individual documents has been proposed (see, for example, Patent Document 1). When the information search apparatus determines that the frequency of use of the search query is high, the information search apparatus uses the statistical information calculated from the search log including the search query to use one or more additional keywords for the search query. Identify refinement queries to include. The information search device generates information that is assumed to be desired by the user using the specified refinement query.

JP 2014-99062 A

However, conventional devices sometimes cannot provide information with high utility value for users.
In view of the above-mentioned circumstances, the present invention aims to provide a technology capable of providing information having higher utility value to users.

One aspect of the present invention includes item identification information which is identification information of an item whose value changes in time series, document identification information which is identification information of a document including at least one sentence, and the document. The specified word number acquisition which acquires the specified word number corresponding to the specific item identification information from the information table by referring to the information table associated with the specified word number indicating the number of specified words which is the evaluation expression specified in advance. And a time-series fluctuation of the value corresponding to the item identification information, and extracting from the information table the number of pieces of document identification information including the input search word associated with the specific item identification information A correlation degree acquisition unit acquiring time series information of the number of pieces of document identification information including the extracted search term and the time series variation; a correlation degree acquired by the correlation degree acquisition unit; Word count On the basis of the obtained specified number of words by parts, an information processing system and an index value acquisition unit that acquires an index value indicating the degree of association between the keyword and the item identification information.

  One aspect of the present invention is the above information processing system, wherein the index value acquisition unit acquires an index value indicating an influence of the search term on the time-series fluctuation.

  One aspect of the present invention is the information processing system, wherein the information table includes sentence information that is a sentence associated with the document identification information, a specified number of words for each sentence information, and the document identification. The date associated with the information is stored, and the correlation degree acquisition unit obtains the number of the document identification information including the search word for each day based on the date associated with the document identification information, Correlation between the number of the document identification information and the time series fluctuation is a value based on the degree of coincidence between the date on which the number of document identification information is greater than or equal to the threshold and the date on which the fluctuation value of the time series fluctuation value is a predetermined value or more Get as a degree.

  One aspect of the present invention is the information processing system described above, wherein the correlation degree acquisition unit is configured such that a date when the number of document identification information is greater than or equal to a threshold value and a variation value of the time-series variation value is greater than or equal to a predetermined value. One or both of the dates are shifted to obtain a value based on the degree of matching as the degree of correlation between the number of document identification information and the time-series fluctuation.

  One aspect of the present invention is the above-described information processing system, wherein the correlation degree acquisition unit associates the document identification information associated with a date prior to the date on which the number of document identification information is acquired. The threshold value is acquired based on the number of.

One embodiment of the present invention is the information processing system described above, wherein the designation word is a positive word indicating a positive expression, or a negative word indicating a negative expression, and the index value acquisition unit When the designated word is a positive word, the positive of the search word and the item identification information is based on the correlation degree acquired by the correlation degree acquiring unit and the number of positive words acquired by the designated word number acquiring unit. The index value indicating the degree of association having a meaningful meaning is acquired, and when the designated word is a negative word, the degree of correlation acquired by the correlation degree acquiring unit and the number of negative words acquired by the designated word number acquiring unit Based on this, an index value indicating a degree of association having a negative meaning between the search term and the item identification information is acquired.

According to one aspect of the present invention, there is provided a computer- implemented item identification information which is identification information of an item whose value changes in time series, document identification information which is identification information of a document including at least one sentence, and the document Refers to the information table associated with the number of designated words indicating the number of designated words which are the evaluation expressions designated in advance, and acquires the number of designated words corresponding to the specific item identification information from the information table The designated word number acquiring step, the time-series fluctuation of the value corresponding to the item identification information, the number of the document identification information including the input search term associated with the specific item identification information, and the information table Extracting a correlation degree between time series information of the number of document identification information including the extracted search term and the time series fluctuation, and the correlation degree acquiring step An index value acquisition step of acquiring an index value indicating a degree of association between the search term and the item identification information based on the acquired correlation degree and the designated word number acquired by the designated word number acquiring step; It is an information processing method.

According to one aspect of the present invention, there is provided a computer comprising : item identification information which is identification information of an item whose value changes in time series; document identification information which is identification information of a document including at least one sentence; Refers to the information table associated with the number of designated words indicating the number of designated words which are the evaluation expressions designated in advance, and acquires the number of designated words corresponding to the specific item identification information from the information table The designated word number acquiring step, the time-series fluctuation of the value corresponding to the item identification information, the number of the document identification information including the input search term associated with the specific item identification information, and the information table extracted from the correlation acquisition step of acquiring correlation between the time-series variation and time-series information of the number of document identification information including the search word and the extracted, by the correlation acquisition step Performing an index value acquisition step of acquiring an index value indicating a degree of association between the search term and the item identification information based on the acquired correlation degree and the number of designated words acquired by the designated word number acquiring step; Is a computer program to

  According to the present invention, it is possible to provide information having higher utility value for the user.

It is a figure for demonstrating the outline | summary of the process which the information processing system 10 performs. It is a schematic block diagram which shows the function structure of the information processing system 10 in 1st Embodiment. It is a figure which shows an example of the item sentence table 25 memorize | stored in the item sentence memory | storage part 24. FIG. 5 is a flowchart showing a flow of processing executed by the information processing system 10. It is a flowchart which shows the flow of the processing detail of step S112. It is a conceptual diagram for demonstrating calculation of a time-sequential correlation score. It is a figure for demonstrating the function structure of 10 A of information processing systems of 2nd Embodiment. It is a figure which shows an example of the information memorize | stored in the item related word list memory | storage part. 5 is a flowchart showing a flow of processing executed by the information registration unit 40.

  Hereinafter, embodiments of an information processing system, an information processing method, and a computer program according to the present invention will be described with reference to the drawings.

First Embodiment
FIG. 1 is a diagram for explaining an outline of processing performed by the information processing system 10. For example, when “Olympics” is input as a search term, the information processing system 10 searches a database constructed based on text information such as news and blog. The information processing system 10 calculates an association index indicating a relationship between a predetermined item (item) and a search term based on the search result. An item refers to a product name, stock item, etc. that is expected for output of search results. In the present embodiment, as an example, a case where a stock brand related to a search word is searched will be described.

  In addition, the information processing system 10 has a positive score (P score) indicating that the search term is determined to have a positive meaning with respect to a predetermined stock name, or the search term denies a predetermined stock name. A negative score (N score) indicating that it is determined to have a specific meaning is calculated. In the case of a stock issue, having a positive meaning means that the issue is positively referred to in the document, or the number of articles that mention the issue is correlated with the rising trend of the stock price. To have a negative meaning means that the relevant stock is mentioned negatively in the document, or the number of articles that mention the relevant stock correlates with the downward trend of the stock price. The information processing system 10 outputs the calculated related index, P score, or N score together with incidental information about stock issues, documents used for search, and the like. The positive score (P score) may be an index value indicating that a search term positively affects a predetermined stock name. The negative score (N score) may be an index value indicating that a search term negatively affects a predetermined stock name.

  The information processing system 10 acquires a search request including a word whose relationship is to be checked from the external terminal 100 via a communication network such as a network. The information processing system 10 searches information stored in an item sentence storage unit 24 (described later) and an item time series storage unit 26 (described later) included in the own device, and is a stock issue that is highly related to a word to be examined. A list of (items) (output item list) is generated and transmitted to the external terminal 100.

  The degree of association is represented by five types of scores: a search score, a positive word score, a negative word score, a positive time series correlation score, and a negative time series correlation score. As the processing result of the information processing system 10, one of five types of scores may be used, or a plurality of types of scores may be combined, and a new value may be calculated by weighted sum of scores or the like. The details of the five types of scores will be described later.

  FIG. 2 is a schematic block diagram illustrating a functional configuration of the information processing system 10 according to the first embodiment. The information processing system 10 is connected to the external terminal 100 via the network NW. The information processing system 10 is configured of one or more information processing apparatuses. For example, when the information processing system 10 includes one information processing apparatus, the information processing apparatus includes a CPU (Central Processing Unit), a memory, an auxiliary storage device, and the like connected by a bus, and executes an information processing program. To do. By the execution of the information processing program, the information processing apparatus has a request reception unit 12, a control unit (index value acquisition unit) 14, a search score acquisition unit 16, a positive gas core acquisition unit (specified word number acquisition unit) 18, and a time It functions as an apparatus including a serial correlation score acquisition unit (correlation degree acquisition unit) 20, a response unit 22, an item sentence storage unit 24, and an item time series storage unit 26. Note that all or part of the information processing system 10 may be realized using hardware such as an application specific integrated circuit (ASIC), a programmable logic device (PLD), or a field programmable gate array (FPGA). Further, the information processing system 10 may be realized by dedicated hardware. The information processing program may be stored in a computer readable storage medium. The computer-readable storage medium is, for example, a portable medium such as a flexible disk, a magneto-optical disk, a ROM, a CD-ROM, a semiconductor storage device (for example, SSD: Solid State Drive), a hard disk built in a computer system, or a semiconductor storage. A storage device such as a device. The information processing program may be provided via a telecommunication line. In the present embodiment, acquisition means that the information processing system 10 or the above-described functional unit of the information processing apparatus is, for example, from another system, other apparatus, other functional unit, etc. via a communication network such as a network. Including obtaining information. In the present embodiment, acquisition includes that the information processing system 10 or the above-described functional unit of the information processing apparatus acquires corresponding information from an information table registered in advance. Furthermore, in the present embodiment, acquisition includes performing arithmetic processing based on the acquired information or information held by the information processing system 10 or the information processing apparatus.

  The request receiving unit 12 acquires a search request including a word (search word) whose relation is to be checked from the external terminal 100 and outputs the search request to the control unit 14. The search request includes, for example, search terms, search target document types (news, blogs, documents related to other companies, or a combination thereof), score calculation method (search score acquisition unit 16, positive / negative score acquisition unit 18, or time series) Among the calculation results calculated by the correlation score acquisition unit 20, which calculation result is finally output or an integrated score based on the calculation result is specified) is included.

  In addition, in search request, output stock list sort order (specify which score to sort based on search score, positive word score, negative word score, positive time series correlation score, negative time series correlation score), output format Etc. are included. The search request includes a search term (the purpose is to search for items associated with this term). In addition, the search request can include the type of target data (news, blog, financial statements, company website, text data related to other brands) and search target. Further, the search request can include parameters (described later) used in the search score acquisition unit 16, the positive gas core acquisition unit 18, or the time-series correlation score acquisition unit 20, instructions regarding information to be output, and the like.

  The item sentence storage unit 24 stores an item sentence table (information table) 25. The item sentence table 25 is created based on text data (hereinafter referred to as “document (document)”) such as news, blog, and other articles. Each record of the item sentence table 25 has a sentence (sentence) which is a unit obtained by dividing the document as a registration unit. FIG. 3 is a view showing an example of the item sentence table 25 stored in the item sentence storage unit 24. As shown in FIG. Corresponding to each sentence, record ID, data type, date associated with the document (available for browsing), stock code related to the sentence, article ID including the sentence, article link First, the article number of the sentence in the document, the body text of the sentence, the search index, the number of positive words, and the number of negative words are associated and stored. The number of positive words is the number of words indicating preset positive expressions present in a sentence. The number of negative words is the number of words indicating preset negative expressions present in a sentence.

The item sentence table 25 is created in advance according to the following procedure, for example, using a collection of documents as input data, and is stored in the item sentence storage unit 24. The item sentence table 25 may be created by the information processing system 10 according to the following procedure, or may be created by the information registration unit 40 of the second embodiment described later.
(1) Sentences are created by dividing a document using punctuation marks, periods, or consecutive line breaks as delimiters.
(2) A well-known text analysis method is applied to each sentence, and a list is created by dividing it into finer units such as morphemes to which part-of-speech information is added. Hereinafter, this unit is called "word". A "word" does not have to be identical to a word or morpheme in linguistics. As a text analysis method, for example, known techniques such as morphological analysis JTAG described in the following reference 1 can be used. In addition, words that are determined not to be useful for searching for particles, auxiliary verbs, and designated pronouns in this processing may be deleted from the list.
[Reference 1] Kenji Imamura and two others, “Basic Japanese Analysis Technology as a Foundation for Knowledge Extraction from Texts”, NTT Technical Journal, v.l.20, No.6, p.20-23, 2008 ( 3) A search index is created using the word created in (2) above.
(4) An evaluation word extraction method is applied to a sentence or a list of words, and an evaluation word label is assigned to each word. For example, two types of labels, for example, "positive" and "negative" are given based on whether they are positive words or negative words. For this processing, for example, a known evaluation expression extraction method described in Reference Document 2 below can be used. Hereinafter, description will be made assuming two types of evaluation word labels, “positive” and “negative”.
[Reference 2] Hisako Asano and three others, “Reputation information indexing technology for analyzing reviews on the Web”, NTT Technical Journal, v.l.20, No.6, p.12-15, 2008 (5) above (4) The label type and the sentence given in (4) are summed up for each label type, and the processing result described above is stored in the item sentence storage unit 24.
(6) The brand code is a code for specifying a brand related to a sentence, and is specified based on the code table stored in the item sentence storage unit 24. In the code table, brands and codes for identifying brands are associated with each other. For example, when a brand in the sentence is included, a brand code in the corresponding sentence is associated.

  The item time-series storage unit 26 stores item time-series information (time-series fluctuation) in which fluctuations in stock prices corresponding to stock issues are associated in time series. When the item is a product name, it is the number of sales per day of the product.

  The control unit 14 analyzes the acquired search request and extracts information corresponding to the analysis result from the item sentence storage unit 24. The control unit 14 outputs the extracted information to the search score acquisition unit 16, positive gas core acquisition unit 18, or time series correlation score acquisition unit 20. The control unit 14 calculates the degree of correlation between various pieces of information based on the calculation results calculated by the search score acquisition unit 16, the positive gas core acquisition unit 18, or the time series correlation score acquisition unit 20. In addition, the control unit 14 is based on the degree of correlation calculated by the time-series correlation score acquiring unit 20 and the calculation result calculated by the positive / negative score acquiring unit 18 in how positive the search word and the stock name are. An index value indicating whether there is relevance or relevance in a negative sense is calculated.

  The search score acquisition unit 16 calculates the search result as a score based on the information extracted by the control unit 14. The positive / negative score acquisition unit 18 determines a positive word score or a negative word for each stock based on the number of words representing a positive expression or the number of words representing a negative expression included in the information extracted by the control unit 14. Calculate the score. The time series correlation score acquisition unit 20 acquires item time series information. The time-series correlation score acquisition unit 20 is based on the item time-series information and the information included in the item sentence table 25, the time-series information (search term time-series information) of the number of documents including the search word and the item time-series information The degree of correlation with is calculated. The response unit 22 transmits the information acquired from the control unit 14 to the external terminal 100 via the network NW.

  FIG. 4 is a flowchart showing a flow of processing executed by the information processing system 10. First, the control unit 14 stands by until a search request is acquired from the request receiving unit 12 (step S100). When the search request is acquired, the control unit 14 analyzes the acquired search request (step S102), and extracts information corresponding to the analysis result from the item sentence storage unit 24 (step S104). Next, the control part 14 extracts the record containing the same brand code based on the extracted information (step S106).

  Next, in the processing from step S108 to step S112, each processing is executed for each identical brand code extracted in step S106. The search score acquisition unit 16 calculates a search score for the brand code based on the number of sentences corresponding to the record extracted in step S106 (step S108). The search score is calculated based on the number of sentences including the search word and associated with the same brand code. The search score corresponds to the related index shown in FIG.

ここでλは実定数であり、検索スコア取得部１６の記憶領域に事前に設定値として保存しておく。なお、式（１）代えて、その他の累積分布関数を用いてもよい。また、検索スコア取得部１６は、単純にセンテンス数をカウントする代わりに、センテンス数に対して重み付けを行ってもよい。例えば検索スコア取得部１６は、センテンスに対応付けられた日付が検索スコアの算出時に近いセンテンスほど大きな重みをかけて和をとるようにしてもよい。また、ニュースやブログなど種別の異なる文書からのセンテンス情報がアイテムセンテンステーブルに含まれる場合、種別ごとに別々に検索スコアを算出し、予め決められた重みを用いて種別ごとの検索スコアの重みづけ和として最終的な検索スコアを決定してもよい。 Although the above sentence number can be used as a search score, for example, assuming that the average value of the extracted sentence numbers follows an exponential distribution of 1 / λ and the standard deviation is 1 / λ, using the cumulative distribution function of the exponential distribution For example, the value of the search score may be calculated using equation (1) so as to satisfy the value between 0 and 100.

Here, λ is a real constant, and is stored in advance in the storage area of the search score acquisition unit 16 as a set value. Note that other cumulative distribution functions may be used instead of equation (1). Further, the search score acquiring unit 16 may weight the sentence number instead of simply counting the sentence number. For example, the search score acquisition unit 16 may add a weight to a sentence whose date closer to the sentence is closer to the time of calculation of the search score, and may sum them. In addition, when sentence information from documents of different types such as news and blogs is included in the item sentence table, the search score is calculated separately for each type, and the search score is weighted for each type using a predetermined weight. The final search score may be determined as the sum.

  Next, the positive / negative score acquisition unit 18 calculates the positive word score Tp and the negative word score Tn based on the number of positive words and the number of negative words included in the sentence of the same grade record extracted in step S106 ( Step S110). The positive word score Tp is calculated based on the count number Cp obtained by adding the number of positive words for each brand code. The negative word score Tn is calculated based on the count number Cn obtained by adding the number of negative words for each brand code. For positive word score Tp or negative word score Tn, the values of Cp and Cn may be used as they are, but similar to the search score, normalized numerical values are calculated using a function such as the equation (1) described above. May be.

  Next, the time series correlation score acquisition unit 20 calculates a time series correlation score (positive score and negative score) (step S112). Details of the method for calculating the time series correlation score, the positive score, and the negative score will be described later. Next, the control unit 14 shapes the calculation result calculated in steps S108 to S112 (step S114). For example, when the sort order of the multiple brand codes output in the search request is designated, the control unit 14 sorts the brand list based on the calculated scores and shapes the list into a predetermined format. The predetermined format is a format specified in a search request such as JSON format, XML format, CSV format and the like. Next, the response unit 22 transmits, to the external terminal 100, the brand list of the predetermined format shaped by the control unit 14 in step S114 (step S116). As a result, the brand list generated by the information processing system 10 is displayed on the external terminal 100. The brand list includes a brand code (brand name), a search score, a positive word score, a negative word score, a time series correlation score to be described later, a positive score to be described later, a negative score to be described later, a searched document name, a URL document, and the like. The information list of, all or part of the ranking is included.

[Method of calculating time series correlation score, positive score, and negative score]
The time-series correlation score acquisition unit 20 is based on the search word time-series information acquired based on the information extracted from the item sentence storage unit 24 and the item time-series information stored in the item time-series storage unit 26. The time series correlation score is calculated. The search term time-series information is information indicating time-series fluctuation of items extracted from the item sentence table 25 by the control unit 14. The search word time series information is, for example, an array of appearance frequencies of search words for each day (described later “document count value time series information”).

  The time series correlation score is a degree indicating the degree of association between the search word time series information and the item time series information. The time series correlation score includes, for example, a positive time series correlation score Sp and a negative time series correlation score Sn. The positive time series correlation score Sp is a score indicating a positive relation in the degree of association between the search word time series information and the item time series information. The negative time series correlation score Sn is a score indicating a negative association in the degree of association between the search word time series information and the item time series information.

  In general, when obtaining a time series correlation score, a Pearson product-moment correlation coefficient between search word time series information and item time series information may be simply obtained. In this case, although the overall tendency of the frequency of appearance of search terms is similar, the case is taken into consideration, for example, when the stock price rises or falls immediately after the day when a large number of search terms appear. It will not be a time series correlation score.

As another method, for example, there is a technique of giving a score to a stock based on the relevancy of the stock price and the text data such as an article described in the following reference 3. In this method, a score is assigned to each word based on the degree of contribution of each word included in the document to the stock price fluctuation. When calculating a time-series correlation score by this method, it is necessary to calculate and hold in advance statistical analysis results such as principal component analysis and regression analysis for all words, and recalculation is performed each time an article is added. It will be necessary. For this reason, since it is not what gives a score directly to items, such as a required brand code, this method may not be able to be used for comparison between brand codes.
[Reference 3] Site group, Kizumi Izumi, "Analysis of stock market trends using multiple text information", Japan Society for Population Intelligence, SIG-FIN-013-09, 2014

  In this embodiment, the time series correlation score acquisition unit 20 calculates a time series correlation score by the following method. Here, the item represents the stock code of the stock. The information stored in the item sentence storage unit 24 is created from a news article including a date and a document. The information stored in the item time series storage unit 26 is the daily stock price (close price, etc.) accumulated for each stock code of stock. The item is not limited to a stock code, but may be an item whose value corresponding to the item varies from day to day. Moreover, the information memorize | stored in the item time series memory | storage part 26 should just be time series data which has the unit same as the unit of the date (or time) information of an item sentence table.

  FIG. 5 is a flowchart showing a detailed flow of processing in step S112. FIG. 6 is a conceptual diagram for explaining the calculation of the time series correlation score.

  First, the time-series correlation score acquiring unit 20 extracts the stock price corresponding to the brand code for which the positive word score Tp and the negative word score Tn are calculated in step S110 (step S150). In this case, the time-series correlation score acquisition unit 20 extracts the stock price on the business day immediately before the period Ds (hereinafter referred to as “stock price correlation value calculation period”) that is initially set. The time-series correlation score acquisition unit 20 may extract the stock prices of all the brand codes stored in the item time-series storage unit 26. Further, the stock price correlation value calculation period may be set based on information included in the search request transmitted from the external terminal 100. The stock price correlation value calculation period is, for example, one year retroactive from the day before the search request issuance date.

  Next, the time-series correlation score acquisition unit 20 calculates two types of date sets, a stock price rise date set Gp and a stock price fall date set Gn, based on the extracted stock price (step S152). The stock price rise date set Gp is a date on which s +% or more rises compared to the previous day (or the previous market business day). The stock price fall date set Gn is a day when the stock price falls by s-% or more compared to the previous day (or the previous market business day). The time-series correlation score acquisition unit 20 sets, for example, 2% as the values of s + and s−. In order to remove the effects of fluctuations in the stock market as a whole, the time series correlation score acquisition unit 20 uses a set value of s +% and s-% as an increase / decrease rate of stock market indicators (eg, Nikkei Average, TOPIX, etc.) The rates may be added dynamically.

  The time-series correlation score acquisition unit 20 calculates a set of document appearance dates Wp for the target stock stock based on the search result acquired by the control unit 14 (step S154). The time-series correlation score acquisition unit 20 calculates, for each day, the number of different documents associated with the sentence for the sentence including the search word acquired from the item sentence storage unit 24 in S104. You may simply use the daily number of sentences. The time-series correlation score acquisition unit 20 calculates the number of documents corresponding to the stock code for Ds + Dw days, which is the sum of the stock price correlation value calculation period Ds and the immediately preceding Dw date, for each day. Hereinafter, the number of documents (or the number of sentences) calculated by the time-series correlation score acquisition unit 20 is referred to as “document count value time-series information”. Further, the time-series correlation score acquisition unit 20 calculates a document appearance date set Wp configured by dates exceeding the threshold value Θ in the stock price correlation value calculation period Ds.

  As the value of Θ, a preset value may be used, but it may be dynamically determined by the following method in order to remove the influence when the number of articles changes depending on the time. That is, for each date of the stock price correlation value calculation period, the time-series correlation score acquisition unit 20 counts the number of documents (sentences) included in the document count value of the past Dw days (for example, Dw day in FIG. The threshold Θ is obtained based on the number. For example, when it is determined whether the day indicated by “X” in FIG. 6 corresponds to a date exceeding the threshold Θ, the threshold Θ is determined based on “Past Dw # day” in the figure. As described above, the document appearance date set Wp can be appropriately calculated by calculating the threshold value by setting the past date immediately before the target date for calculating the number of documents (the number of sentences) as the Dw date.

The threshold Θ can be calculated, for example, using equation (2) with μ as an average of document counts in the past Dw days and σ as a standard deviation. Here, the value of α is a value set in advance, for example, α = 1.0.

ここで、Ｇｘは、株価上昇日集合Ｇｐまたは株価下落日集合Ｇｎを示す。Ｓｘは、Ｇｘに対応したポジティブ時系列相関スコアＳｐ、またはネガティブ時系列相関スコアＳｎを示す。絶対値記号は各集合の要素数を表す。Ｇｘ∩Ｗｐは、ＧｘとＷｐが共通して含む日付からなる集合である。 The time-series correlation score acquisition unit 20 calculates a positive time-series correlation score Sp based on the matching degree between the document appearance date set Wp and the stock price rise date set Gp (step S156). Further, the time series correlation score acquisition unit 20 calculates a negative time series correlation score Sn based on the degree of match between the document appearance date set Wp and the stock price fall date set Gn (step S158). The time-series correlation score acquiring unit 20 calculates the positive time-series correlation score Sp and the negative time-series correlation score Sn, for example, using a die coefficient or the like as shown in equation (3).

Here, Gx indicates a stock price rise date set Gp or a stock price fall date set Gn. Sx indicates a positive time-series correlation score Sp corresponding to Gx, or a negative time-series correlation score Sn. The absolute value symbol represents the number of elements in each set. Gx ∩ Wp is a set of dates which Gx and Wp contain in common.

  The time-series correlation score acquisition unit 20 determines whether or not to calculate a score obtained by applying a time lag in which the document affects the item time-series information (step S160). The number of days of lag to be applied may be set in advance, and S160 may be omitted. The above-described formula (3) gives a high score to a pair of a brand and a search word having a high common ratio between the document appearance date and the stock price (time series) when the document rises or falls. If the stock price fluctuates one day or two days after the document appearance date and if you are interested in a pair of search terms, for example, based on an instruction transmitted from the external terminal 100, the time-series correlation score acquisition unit 20 Calculate the score applying the time lag of

  When it is determined to calculate the score to which the time lag is applied, the time series correlation score acquisition unit 20 calculates the positive time series correlation score Sp or the negative time series correlation score Sn to which the date time lag is applied (step S162). Specifically, the time-series correlation score acquisition unit 20 calculates a score focusing on the common part of the document appearance date before i days and the stock price rise date / fall date. The time-series correlation score acquisition unit 20 creates a new document appearance date set Wpi in which each date included in the document appearance date set Wp is shifted i days later for i = 1, 2,. The time series correlation score acquisition unit 20 calculates the positive time series correlation score Spi or the negative time series correlation score Sni by applying the above-described equation (3) to each Wpi.

ここで、Ｓｘ’は、ポジティブ時系列相関スコアＳｐまたはネガティブ時系列相関スコアＳｎを表し、ＳｘｉはＳｐｉまたはＳｎｉのいずれかを表している。Ｓｐ０＝Ｓｐ、Ｓｎ０＝Ｓｎとする。また、ｗ_ｉは、予め設定された値である。なお、ドキュメント出現日集合Ｗｐｉの作成に代えて、株価上昇日集合Ｇｐまたは株価下落日集合Ｇｎに含まれる各日付をｉ日前にずらした株価上昇日集合Ｇｐｉまたは株価下落日集合Ｇｎｉを作成してもよい。 Then, the time-series correlation score acquisition unit 20 performs weighted sum using Equation (4), and calculates a final positive time-series correlation score Spi or a negative time-series correlation score Sni.

Here, Sx 'represents a positive time-series correlation score Sp or a negative time-series correlation score Sn, and Sxi represents either Spi or Sni. It is assumed that Sp0 = Sp and Sn0 = Sn. W _i is a preset value. It should be noted that instead of creating the document appearance date set Wpi, a stock price rise date set Gpi or a stock price fall date set Gni is generated by shifting each date included in the stock price rise date set Gp or the stock price fall date set Gn i days ago. It is also good.

  Next, the control unit 14 calculates a positive score Ip and a negative score In (step S164). The control unit 14 acquires a calculation result for each brand from the search score acquisition unit 16, positive gas core acquisition unit 18, and time-series correlation score acquisition unit 20. The control unit 14 acquires a list of search scores from the search score acquisition unit 16. The control unit 14 acquires a list of positive word scores and negative word scores from the positive gas core acquisition unit 18. The control unit 14 acquires a list of positive time series correlation scores and negative time series correlation scores from the time series correlation score acquisition unit 20.

ここで、Ｃｘは、ポジティブ語カウント数Ｃｐまたはネガティブ語カウント数Ｃｎを表す。βは、予め設定されたパラメータであり、例えばβ＝１．０または２．０等の値である。 The control unit 14 integrates the results of the positive / negative score acquisition unit 18 and the time-series correlation score acquisition unit 20, and integrates indices into one for positive and negative. The positive word count number Cp and the positive time-series correlation score Sp are integrated to calculate a positive score Ip. The control unit 14 integrates the negative word count number Cn and the negative time-series correlation score Sn to calculate a negative score In. The control unit 14 applies, for example, the formula (1) by regarding the converted count number as the count number, using, for example, the converted count number calculated using the equation (5), to obtain the positive score Ip (P score ) And negative score In (N score).

Here, Cx represents a positive word count number Cp or a negative word count number Cn. β is a preset parameter, for example, a value such as β = 1.0 or 2.0.

  According to the information processing system 10 of the first embodiment described above, the correlation between the appearance degree of the search word calculated by the time series correlation score acquisition unit 20 and the fluctuation of the stock price is calculated by the positive gas core acquisition unit 18. Based on the number of positive words (or the number of negative words), the index value indicating the strength of the relationship between the search term and items such as stocks, etc. Since the calculation is performed for each context, it is possible to provide information having high utility value for the user.

Second Embodiment
The second embodiment will be described below. In the second embodiment, the information registration unit 40 acquires information for calculating each score from the article provider 60 and the time-series data provider 70, and stores the information in the item sentence storage unit 24 or the item time-series storage unit 26. It differs from the first embodiment in the points. Hereinafter, this difference will be mainly described.

  FIG. 7 is a diagram for describing a functional configuration of an information processing system 10A according to the second embodiment. The external terminal 100 is communicably connected to the user terminals 110, 120, and 130 via the network NW. In the second embodiment, for example, the external terminal 100 receives a search request from the user terminal 110, 120, or 130 and transmits the received search request to the request receiving unit 12. In addition, the external terminal 100 acquires the processing result processed by the information processing system 10 </ b> A from the response unit 22. The external terminal 100 generates the acquired processing result as a predetermined image including the processing result, and transmits it to the user terminal 110, 120, or 130.

  The information processing system 10A further includes an information registration unit 40 in addition to the functional configuration of the information processing system 10 of the first embodiment. The information registration unit 40 includes an item sentence registration unit 42, an item extraction unit 44, an item related word list storage unit 46, and an item time series information registration unit 48.

  The item sentence registration unit 42 obtains a document of an article (news, blog, etc.) described in text via the network NW from an external article provider 60 that creates, collects, and distributes articles. The item sentence registration unit 42 extracts necessary information from the acquired document, and stores the extracted necessary information in the item sentence storage unit 24 in units of sentences. The necessary information includes, for example, a date, a link to an article such as a URL, a title, an article text, and the like.

  The item extraction unit 44 includes one or more related words (hereinafter referred to as “item related words”) stored in an item related word list storage unit 46 described later in the document acquired by the item sentence registration unit. Determine if it is included in the document. The item extraction unit 44 outputs an item list in which items included in each document are listed based on the determination result. This process may use a generally used character string search technique, or may use a search index generated based on morphemes obtained by morphological analysis of a document. For example, as a character string search technique, a grep command that is standardly used in a UNIX (registered trademark) computer may be used.

As an item, for example, when a stock issue is considered, the issue name may be expressed as the same character string as another proper noun such as a general noun or a place name. In such a case, in order to avoid erroneous extraction of items, it is a method of giving labels such as "organization name", "place name", "person's name", "date" to words and morphemes included in a document The item extraction unit 44 may extract the item only when applying the specific expression extraction and the related term of the item is labeled “organization name”. For the proper expression extraction, for example, known techniques such as those in Reference 1 and Reference 4 below can be used.
[Reference 4] Hideki Kashiwazaki, Hideto Kawazawa, "Speeding up SVM for Named Entity Extraction", Transactions of Information Processing Society of Japan, vol. 44, No. 3, p. 970-979, 2003

  The item related word list storage unit 46 stores an item name (brand) and a related word related to the item name in association with each other. FIG. 8 is a diagram illustrating an example of information stored in the item-related word list storage unit 46. For example, when the item name is a company name, an abbreviation of the company name, a product or service provided by a company corresponding to the company name, and the like are stored as item-related words.

  The item time-series information registration unit 48 obtains time-series data via the network NW from an external time-series data provider 70 that creates or collects and distributes time-series data regarding items. For example, when the item is a stock brand, the time-series data is information including a stock price at each predetermined time or every predetermined time of each brand.

  FIG. 9 is a flowchart showing a flow of processing executed by the information registration unit 40. First, the item sentence registration unit 42 acquires a document list from the article provider 60, and reads the acquired document list (step S200). Next, the item sentence registration unit 42 determines whether or not a document is included in the read document list (step S202). If the document is not included, the processing of this flowchart ends.

  When a document is included in the document list, the item extraction unit 44 extracts a related item from the document and generates an item list (step S204). Next, the item extraction unit 44 determines whether or not there is one or more item list generation (step S206). If the generated item list does not exist, the process of this flowchart ends.

  When one or more item lists are generated, the item sentence registration unit 42 divides the document into sentence units based on periods, punctuation marks, and continuous line breaks (step S208).

  Next, the item sentence registration part 42 performs the process of step S210 to step S214 for every divided sentence. The item sentence registration unit 42 divides the sentence into “words” using a known morphological analysis technique, and generates a search index using each word as a key (step S210). When the item extraction unit 44 generates a search index in step S204, the search index generated by the item extraction unit 44 may be used.

  Next, the item sentence registration unit 42 extracts positive words and negative words, and calculates the number of morphemes or words indicating positive expressions and negative expressions, respectively (step S212). The item sentence registration unit 42 may extract a positive expression or a negative expression based on, for example, a list of positive expressions (for example, “beautiful”, “improvement”,...) Prepared in advance. A known evaluation expression extraction technique as described in the above may be used. At this time, the item sentence registration unit 42 may register, for example, the number of extracted positive expressions or negative expressions in the item sentence storage unit 24. Next, the item sentence registration unit 42 generates a list similar to the item sentence table 25, and registers the generated list in the item sentence storage unit 24 (step S214). Thus, the processing of this flowchart ends.

  The process executed by the information registration unit 40 described above is a method in which the same item list is always assigned to sentences included in the same document. However, there are cases where the brands are completely different from the first half and second half of the document. Even in this case, the item list for each sentence can be narrowed down by the following method.

(1) In each sentence, an item which does not include an item related term in the sentence generates a temporary item list excluded from the item list (only an item containing an item related term in each sentence is left).
(2) After applying the above (1) to the sentences in each document, in each sentence, combine the temporary item lists for the sentences of the nl records before and after that, and register the ones without duplicates as the new item list ( Use only items contained within the nl record before and after)). Note that nl is set in advance. Further, the item sentence registration unit 42 may periodically or at the time of registration process delete a sentence whose storage period (for example, one year) set in advance has passed.

  In the information processing system 10A of the second embodiment described above, the item sentence registration unit 42 performs morphological analysis, extraction of positive words and negative words, and the like based on the document list acquired from the article provider 60, and then the items A list similar to the sentence table 25 is generated. As a result, information used by the information processing system 10 to calculate the index value can be appropriately generated.

  In the present embodiment, the item is described as a stock issue as an example, but the item is not limited to the stock issue. For example, the item only needs to be one in which the value given to the item varies in time series. For example, the item may be a product, a service, or an index whose price varies. The items may be, for example, petroleum, brand of futures trading, foreign exchange, various securities, inflation rate, television program rating, sales of goods, and the like.

  You may make it implement | achieve the function of the information processing system 10 in embodiment mentioned above with a computer. In that case, a program for realizing this function may be recorded in a computer readable recording medium, and the program recorded in the recording medium may be read and executed by a computer system. Here, the “computer system” includes an OS and hardware such as peripheral devices. The “computer-readable recording medium” refers to a storage device such as a flexible medium, a magneto-optical disk, a portable medium such as a ROM and a CD-ROM, and a hard disk incorporated in a computer system. Furthermore, “computer-readable recording medium” dynamically holds a program for a short time, like a communication line in the case of transmitting a program via a network such as the Internet or a communication line such as a telephone line. It may also include one that holds a program for a certain period of time, such as volatile memory in a computer system that becomes a server or client in that case. The program may be for realizing a part of the functions described above, or may be realized in combination with the program already recorded in the computer system.

  The embodiment of the present invention has been described in detail with reference to the drawings. However, the specific configuration is not limited to this embodiment, and includes design and the like within the scope of the present invention.

10 ... information processing system, 12 ... request reception unit, 14 ... control unit, 16 ... search score acquisition unit,
18: positive / negative score acquisition unit, 20: time series correlation score acquisition unit, 22: response unit, 24: item sentence storage unit, 26: item time series storage unit, 40: information registration unit, 42: item sentence registration unit, 44 ... Item extraction unit, 46 ... Item related word list storage unit, 48 ... Item time series information registration unit

Claims (8)

Item identification information which is identification information of an item whose value changes in time series, document identification information which is identification information of a document including at least one or more sentences, and a previously designated evaluation expression included in the document A designated word number acquisition unit which acquires the designated word number corresponding to specific item identification information from the information table with reference to an information table in which the designated word number indicating the number of designated words is associated;
The time series variation of the value corresponding to the item identification information is acquired, the number of the document identification information including the search word associated with the specific item identification information and including the input search term is extracted from the information table, and the extraction is performed A correlation degree acquisition unit that acquires a degree of correlation between time series information of the number of document identification information including the searched search term and the time series fluctuation;
An index value for acquiring an index value indicating the degree of association between the search term and the item identification information, based on the degree of correlation acquired by the degree of correlation acquiring unit and the number of designated words acquired by the designated word number acquiring unit Acquisition part,
An information processing system comprising: The index value acquisition unit acquires an index value indicating an influence of the search term on the time-series fluctuation;
The information processing system according to claim 1. The information table stores sentence information that is a sentence associated with the document identification information, the number of designated words for each sentence information, and a date associated with the document identification information.
The correlation degree acquisition unit acquires the number of the document identification information including the search term for each day based on the date associated with the document identification information, and the date when the number of the document identification information is equal to or more than a threshold Acquiring a value based on the matching degree between the date and the date when the fluctuation value of the time-series fluctuation value is a predetermined value or more as the correlation degree between the number of the document identification information and the time-series fluctuation.
The information processing system according to claim 1 or 2. The correlation degree acquiring unit shifts one or both of the date in which the number of the document identification information is a threshold or more and the date in which the fluctuation value of the time-series fluctuation value is a predetermined value or more. Acquiring a value based on the degree of correlation between the number of the document identification information and the time-series fluctuation,
The information processing system according to claim 3. The correlation degree acquisition unit acquires the threshold based on the number of the document identification information items associated with a date before the date on which the acquisition of the number of the document identification information objects is performed.
The information processing system according to claim 3 or 4. The designation word is a positive word indicating a positive expression, or a negative word indicating a negative expression,
The index value acquisition unit
When the designated word is a positive word, the affixing of the search term and the item identification information is performed based on the correlation degree acquired by the correlation degree acquiring unit and the number of positive words acquired by the designated word number acquiring unit. An index value indicating the degree of relevance
When the designated word is a negative word, the negation of the search term and the item identification information is performed based on the correlation degree acquired by the correlation degree acquiring unit and the number of negative words acquired by the designated word number acquiring unit. An index value that indicates the degree of relevance
The information processing system according to any one of claims 1 to 5. Computer
Item identification information which is identification information of an item whose value changes in time series, document identification information which is identification information of a document including at least one or more sentences, and a previously designated evaluation expression included in the document A designated word number acquisition step of acquiring from the information table the designated word number corresponding to specific item identification information by referring to an information table in which the designated word number indicating the number of designated words is associated;
The time series variation of the value corresponding to the item identification information is acquired, the number of the document identification information including the search word associated with the specific item identification information and including the input search term is extracted from the information table, and the extraction is performed Acquiring a degree of correlation between time series information of the number of document identification information including the searched search term and the time series fluctuation;
An index value for acquiring an index value indicating the degree of association between the search term and the item identification information based on the degree of correlation acquired in the degree of correlation acquisition step and the number of designated words acquired in the number of designated words acquisition step Acquisition step,
Information processing method having. On the computer
Item identification information which is identification information of an item whose value changes in time series, document identification information which is identification information of a document including at least one or more sentences, and a previously designated evaluation expression included in the document A designated word number acquisition step of acquiring from the information table the designated word number corresponding to specific item identification information by referring to an information table in which the designated word number indicating the number of designated words is associated;
The time series variation of the value corresponding to the item identification information is acquired, the number of the document identification information including the search word associated with the specific item identification information and including the input search term is extracted from the information table, and the extraction is performed Acquiring a degree of correlation between time series information of the number of document identification information including the searched search term and the time series fluctuation;
An index value for acquiring an index value indicating the degree of association between the search term and the item identification information based on the degree of correlation acquired in the degree of correlation acquisition step and the number of designated words acquired in the number of designated words acquisition step Acquisition step,
A computer program to run

Images