JP4800846B2 - Topic degree calculation method and apparatus, program, and computer-readable recording medium - Google Patents

Description

  The present invention relates to a topic level calculation method and apparatus, a program, and a computer-readable recording medium. In particular, in a situation where documents including new information can be obtained one after another, a phrase that is a topic from a document group is automatically detected. The present invention relates to a topic level calculation method and apparatus, a program, and a computer-readable recording medium.

  With the development of the Internet and other information media, anyone can easily send information, and documents are created by various senders and sent on the network. Under such circumstances, it is considered that it becomes possible to extract matters that have become a topic at an arbitrary point in time by analyzing document information created so far.

  A technology that extracts feature words that are a topic in a document group from a document group uploaded on a network system typified by the Internet, taking into account the temporal variation of the number of occurrences of the phrase contained in the document group. Several have been proposed.

  As a conventional technique, a document uploaded on a network system is acquired together with its creation time information, and automatically classified into a plurality of fields set in advance according to the contents of the document. There is a technique for calculating a topic level value indicating a high topic level as a characteristic word representing a topic with respect to a word that has a characteristic increase along the line and that does not appear in other fields (for example, , See Patent Document 1).

However, in the above technique, when the number of times a certain phrase is used during a certain period of time is extracted as a feature phrase that expresses a topic, the total document volume increases when the total document volume increases in the short term. When it fluctuates regularly, there is a problem that the accuracy is lowered due to the influence. For this reason, there is a method of setting a correction function that cancels fluctuations based on changes in the document amount (see Non-Patent Document 1, for example).
JP 2005-276115 A "Discovery of burst in document stream" Yukiaki Fujiki, Yasuyuki Minamino, Yasuhiro Suzuki, Manabu Okumura, IPSJ Research Report 2003-NL-160

  However, the above-described conventional method is troublesome because the user must grasp the pattern of change in the document amount and design the correction function.

  The present invention has been made in view of the above points, and it is possible to automatically correct the influence and extract a topic word / phrase with high accuracy regardless of the variation of the parameter of the number of documents. An object of the present invention is to provide a topic level calculation method, apparatus, program, and computer-readable recording medium.

  FIG. 1 is a diagram for explaining the principle of the present invention.

The present invention (Claim 1) is a topic level calculation method in an apparatus that analyzes a large number of documents and determines the strength of topicality at a desired time with respect to a phrase included in the document,
When an input document group having creation time information is input, the document analysis unit analyzes the input document group and extracts a word / phrase (hereinafter referred to as an in-document word / phrase) to be evaluated. input document number, counts the word frequency is the number of times of use of the document in terms, the document analysis storing the word database (step 110) with their aggregations and aggregation time and the document in the phrase,
The phrase frequency calculation means is a phrase (hereinafter referred to as a processing target phrase w ) that is subject to topic level calculation processing input from the outside, and time information (hereinafter, calculation designation) that indicates the topic at which the topic is calculated. An input receiving step (step 120) for acquiring time)
Phrase frequency calculation means searches a phrase database based on a processing target phrase, and a phrase frequency function representing a temporal variation of the phrase frequency based on a phrase frequency and a total time of a phrase in the document corresponding to the processing target phrase w A phrase frequency calculating step (step 130) for calculating G _w (T) ;
The phrase calculated by the topic frequency calculation step (step 130) is the topic level indicating the level of topicality of the processing target phrase at the time of the input time input at the input reception step (step 120). A topic level calculation step (step 140) to be obtained by multiplying the frequency function G _w (T) by a topic weight function that calculates a high topic level when the phrase frequency is increased near the calculation specified time;
And
In the phrase frequency calculation step (step 130),
The phrase frequency calculation means obtains a time variation D _w (T) of the phrase frequency in the input document of the processing target phrase w and a time variation D _all (T) of the number of input documents, and D _w (T) and D _all The degree of correlation R (w) is obtained by evaluating the similarity of (T), and a value proportional to the value obtained by multiplying the D _all (T) by the R (w) is subtracted from the D _w (T). The obtained value is set as a phrase frequency function G _w (T) .

Further, the present invention (Claim 2 ) provides a phrase frequency calculating step (Step 130).
The correlation degree R (w) between the time fluctuation D _w (T) of the word frequency in the input document group of the processing target word w and the time fluctuation D _all (T) of the number of input documents

により求める。

Ask for.

Further, the present invention (Claim 3 ), in the phrase frequency calculation step (Step 130),
The number of input documents fluctuated using the degree of correlation R (w) between the time fluctuation D _w (T) of the word frequency in the input document of the processing target word w and the time fluctuation D _all (T) of the number of input documents. The phrase frequency function G _w (T) considering the increase or decrease of the phrase frequency of the phrase due to the influence,

により求める。

Ask for.

Further, the present invention (Claim 4 ) is provided in the input receiving step (Step 120).
Instead of accepting the processing target phrase w that is the subject of the topic level calculation process and the calculation designated time indicating the topic level at which to calculate,
In the phrase frequency calculation step (step 130),
The phrase frequency function G _w (T) is obtained for all the processing target phrases included in the input document group,
In the topic level calculation step (step 140),
The topic level at the processing time is obtained for all the processing target words included in the input document group.

  FIG. 2 is a principle configuration diagram of the present invention.

The present invention (Claim 5) is a topic degree calculation device that analyzes a large number of documents and determines the strength of topicality at a desired time for a word or phrase included in the documents,
When an input document group having creation time information is input, the input document group is analyzed to extract a word / phrase (hereinafter referred to as an in-document word / phrase) for topicality evaluation, and the number of input documents input, A document analysis unit 210 that counts the phrase frequency, which is the number of times the phrase in the document is used, and stores it in the phrase database 230 together with the tabulated time and the phrase in the document;
A word / phrase (hereinafter referred to as a processing target word / w ) input from the topic level calculation process inputted from the outside is acquired, the word / phrase database 230 is searched based on the processing target word / phrase, and the acquired processing target word / phrase w A phrase frequency calculating means 240 for calculating a phrase frequency function G _w (T) representing a temporal variation of the phrase frequency based on the phrase frequency and the aggregation time of the corresponding phrase in the document;
Obtaining time information (hereinafter referred to as calculation designated time) indicating the topic at which the topic is calculated from the outside, and at the time of the calculation designated time, the topic degree representing the level of topicality of the processing target phrase, The topic obtained by multiplying the phrase frequency function G _w (T) calculated by the phrase frequency calculation unit 240 by the topic weight function for calculating a high topic degree when the phrase frequency is increasing near the input time. Degree calculation means 250;
Have
The phrase frequency calculation means 240
The time fluctuation D _w (T) of the word frequency in the input document of the processing target word w and the time fluctuation D _all (T) of the number of input documents are obtained, and the similarity between D _w (T) and D _all (T) The degree of correlation R (w) is evaluated, and a value proportional to the value obtained by multiplying the D _all (T) by the R (w) is subtracted from the D _w (T). Including means for G _w (T) .

Further, the present invention (claim 6 ) provides the phrase frequency calculation means 240,
The correlation degree R (w) between the time fluctuation D _w (T) of the word frequency in the input document group of the processing target word w and the time fluctuation D _all (T) of the number of input documents

により求める。

Ask for.

Further, the present invention (claim 7 ) provides the phrase frequency calculating means 240,
The number of input documents fluctuated using the degree of correlation R (w) between the time fluctuation D _w (T) of the word frequency in the input document of the processing target word w and the time fluctuation D _all (T) of the number of input documents. The phrase frequency function G _w (T) considering the increase or decrease of the phrase frequency of the phrase due to the influence,

により求める。
また、本発明（請求項８）は、語句頻度算出手段２４０において、話題度算出処理の対象となる処理対象語句と、どの時点での話題度を算出するかを表す算出指定時刻を受け付ける代わりに、入力文書群中に含まれる全ての処理対象語句について語句頻度関数を求める手段を含み、
話題度算出手段２５０において、入力文書群中に含まれる全ての処理対象語句について処理時刻での話題度を求める手段を含む。

Ask for.
Further, according to the present invention (claim 8 ), the phrase frequency calculation unit 240 accepts a processing target phrase that is a subject of the topic degree calculation process and a calculation designated time that indicates when the topic degree is calculated. , Including means for obtaining a phrase frequency function for all the processing target phrases included in the input document group,
The topic level calculation means 250 includes means for determining the topic level at the processing time for all the processing target words included in the input document group.

The present invention (Claim 9 ) is a topic level calculation program for causing a computer to execute each means of the topic level calculation apparatus according to claims 5 to 8 .

The present invention (Claim 10) is a computer-readable recording medium storing a topic degree calculation program for causing a computer to execute each means of the topic degree calculation apparatus according to claims 5 to 8 .

  As described above, according to the present invention, document information that is sent one after another such as news articles and diary articles published on the web is acquired, and the topical level of words in the document is automatically extracted. In this case, it is possible to remove the influence of the time variation of the total number of documents. Thereby, when the total number of documents according to the prior art fluctuates, an erroneous topic word is not extracted, and it is possible to extract recent trends and topics with high accuracy.

  Hereinafter, embodiments of the present invention will be described with reference to the drawings.

[First Embodiment]
FIG. 3 shows the configuration of the topic level calculation device according to the first embodiment of the present invention.

  The topic level calculation apparatus shown in FIG. 2 is connected to a document database 200 that stores document data that is input to the apparatus and a topic display apparatus 260 that displays topic word information output from the apparatus.

  The topic level calculation device includes a document analysis unit 210, a phrase totaling unit 220, a phrase database 230, a phrase frequency calculation unit 240, and a topic level calculation unit 250.

  The document database 200 stores a document group to which a creation time is added. For example, a document database is created by adding a creation time such as “2006 4/25 13:55” and a document ID uniquely representing a document to a document published on the Web, and inputting and recording them one after another. 200 can be built. In the case of an information source in which new documents are sequentially updated, such as a diary site on the Internet, even when a document in the site is updated, the new document may be regarded as being created and collected.

The document analysis unit 210 acquires documents stored in the document database 200 one document at a time, performs morphological analysis, and disassembles each part of speech. For example, the sentence “delicious chocolate donut” is broken down into “delicious”, “chocolate”, and “donut”. Select and extract only nouns from the decomposed part-of-speech group. At this time, if necessary, consecutive nouns such as “chocolate” and “donut” may be connected to form a compound noun “chocolate donut”, and the compound noun may be treated as one noun. In the following description, nouns and compound nouns are collectively referred to as words. For each word obtained in this way, the creation time and document ID of the document containing the word before analysis are added,
"Choco Donut 2006/01/06 11:36 ID1035, Banana 2006/01/06 11:36 ID1035, New Product 2006/01/06 12:06 ID1036,…"
Is stored in a phrase buffer (not shown) of the phrase totaling unit 220.

The phrase totaling unit 220 is activated at predetermined intervals, reads information stored in the phrase buffer, takes out document IDs included in the information in the phrase buffer without duplication, and sets the number as the document number D _all. aggregated, also counts the number of uses of the word w _k as word frequency Dw _k for all words contained in the word buffer. The total number of documents D _all obtained as a result of the aggregation and a set of all the phrases w _k and the phrase frequencies Dw _k of the phrases w _k are recorded in the phrase database 230 together with information on the total times. An example of information stored in the phrase database 230 is shown in FIG.

In word counting unit 220 for data amount reduction, instead of counting the number of uses of the word w _k, to aggregate once combined if multiple word w _k with the same document ID is used Thus, the phrase frequency Dw _k of the phrase w _k may be aggregated.

  When the phrase information that is subject to topic level calculation is input from the outside, the phrase frequency calculation unit 240 searches the phrase database 230 based on the input phrase and acquires the phrase frequency and the number of documents corresponding to the processing target phrase. Then, by comparing the time changes, the frequency of use of the topic level calculation target words / phrases excluding the influence of the time variation of the total number of documents is stored in a buffer (not shown) of the topic level calculation unit 250. Output.

  FIG. 5 is a flowchart of processing of the phrase frequency calculation unit in the first embodiment of the present invention.

  Step 500) The phrase frequency calculation unit 240 receives an input of phrase information w to be processed from the outside when the process is started.

Step 510) A function that searches the phrase database 230 based on the accepted processing target phrase w, reads usage frequency information for each tabulation period corresponding to the word w, and represents a variation in the usage frequency of the phrase w for each tabulation period D _w (T) is obtained. Here, T is a discrete value. As an example, FIG. 6 shows D _w1 (T), D _w2 (T), and D _w3 (T) representing the usage frequency fluctuations in _three words w ₁ , w _2, and w ₃ . The curve shown in FIG. 6 is actually a set of discrete points. At this time, D _w (T) may be calculated only for the number of documents used in the most recent N period in order to reduce processing.

Step 520) Next, the number of documents for each counting period is acquired from the phrase storage database 230, and the variation D _all (T) of the number of documents for each counting period is calculated. An example of D _all (T) is shown in FIG. Note that the curve shown in FIG. 7 is actually a set of discrete points. At this time, D _all (T) may be calculated only for the number of documents in the latest N periods for reducing processing.

Step 530) The degree of correlation between the frequency of use of the processing target phrase w in the input document group and the number of input documents is calculated as the correlation degree R (w). Specifically, the degree of correlation R (w) is obtained by evaluating the similarity between the time variation function D _w (T) of the phrase w and the time variation function D _all (T) of the number of documents. The similarity is calculated using the following equation which is a general waveform correlation function.

ステップ５４０） 次に、入力文書数の時間毎の変化量に相関度Ｒ（ｗ）を掛け合わせた値に比例した値を処理対象語句の同時間の使用頻度から除くことにより、入力文書数が変動した影響による語句の使用頻度の増減を考慮した語句頻度関数Ｇ_w（Ｔ）を求める。具体的には、Ｄ_all（Ｔ）の最小値をminＤ_all（Ｔ）、最大値をmaxＤ_all（Ｔ）とし、次の式で求める。

Step 540) Next, the value proportional to the value obtained by multiplying the amount of change in the number of input documents with time by the correlation degree R (w) is removed from the frequency of simultaneous use of the processing target words, thereby obtaining the number of input documents. A phrase frequency function G _w (T) that takes into account the increase or decrease in the phrase usage frequency due to the effect of fluctuation is obtained. Specifically, the minimum value of D _all (T) is minD _all (T), and the maximum value is maxD _all (T).

この際、計算の簡略化のため、Ｒ（ｗ）が０よりも小さい場合には、Ｒ（ｗ）＝０とみなして計算してもよい。また、Ｇ_w（Ｔ）が０よりも小さくなる場合には、Ｇ_w（Ｔ）＝０としてもよい。

At this time, in order to simplify the calculation, when R (w) is smaller than 0, it may be calculated assuming that R (w) = 0. When G _w (T) is smaller than 0, G _w (T) = 0 may be set.

In this way, by calculating the change in the phrase usage frequency excluding the influence of the time variation of the total number of documents, the phrase frequency function G _w1 (in which the influence of the total number of documents of each phrase shown in FIG. T), G _w2 (T), and G _w3 (T) are as shown in FIG. The curves and straight lines shown in FIG. 8 are actually a set of discrete points.

Step 550) The phrase frequency function G _w (T) of the phrase w obtained above is recorded in a phrase frequency buffer (not shown) of the topic degree calculation unit 250.

  Next, processing of the topic level calculation unit 260 will be described.

  FIG. 9 is a flowchart of the process of the topic level calculation unit in the first embodiment of the present invention.

Step 910) topics calculating unit 260, the word frequency of word frequency buffer (not shown) function G _w (T) is written, the process was started and receives an input of the processing target time t _p from the outside.

Step 920) Next, based on the processing target time corresponds to the process range up to t _p -t _q from t _p obtained using the positive value t _q previously given, the recently much used words A topic weighting function It _p (t) that creates a large weight is created. At this time, it is preferable to create a weighting function such as the impact curve shown in Patent Document 1 described above.

Step 930) Next, the topic degree TS (w) of the phrase w is obtained by substituting the topic weighting function It _p (t) and the temporal change G _w (T) of the usage frequency of the phrase w into the following equation.

ステップ９４０） 得られた語句ｗの話題度ＴＳ（ｗ）と語句ｗとを表示装置２６０に出力する。例えば、「決勝戦32.8」といった結果が表示装置２６０の画面上に出力される。

Step 940) The topic level TS (w) and the phrase w of the obtained phrase w are output to the display device 260. For example, a result such as “final game 32.8” is output on the screen of the display device 260.

  Note that, in step 500 in the above phrase frequency calculation unit 240 and in step 910 in the topic level calculation unit 260, the phrase information and the processing target time are input from the outside, respectively, but this is limited to this example. Instead, the phrase frequency calculation unit 240 may input both.

[Second Embodiment]
FIG. 10 shows the configuration of the topic level calculation apparatus according to the second embodiment of the present invention.

  The topic level calculation apparatus shown in FIG. 2 is connected to a document database 200 that stores document data that is input to the apparatus, and a topic phrase recording apparatus 1060 that records topic phrase information output by the apparatus.

  The topic level calculation device includes a document analysis unit 210, a phrase totaling unit 220, a phrase database 230, a phrase frequency calculation unit 1040, and a topic level calculation unit 1050.

  Of the above configuration, the document database 200, the document analysis unit 210, the phrase totaling unit 220, and the phrase database 230 are the same as those in the first embodiment described above, and a description thereof will be omitted.

  In the phrase topic level calculation method shown in the first embodiment, the topic level calculation process is performed after receiving an input of a phrase that is a topic level calculation target from the user. This allows you to know if a word is a topic, but not when an unknown word is a topic.

  On the other hand, in this embodiment, by calculating the topic level for all the words and phrases included in the phrase database 230 at predetermined intervals, it is possible to determine what kind of phrase is being talked about at any given time. Extractable.

  Hereinafter, operations of the phrase frequency calculation unit 1040 that calculates the phrase frequency of all words and phrases and the topic level calculation unit 1050 that calculates the topic level of all phrases that are not in the first embodiment will be described.

  FIG. 11 is a flowchart of processing of the phrase frequency calculation unit in the second embodiment of the present invention.

  Step 1100) The phrase frequency calculation unit 1040 is activated every predetermined period, accesses the phrase database 230, takes out the stored phrases without duplication, creates a phrase list, and stores the phrase frequency calculation unit 1040 in the phrase frequency calculation unit 1040 Stored in a buffer (not shown).

Step 1110) Next, the number of documents for each counting period is acquired from the phrase storage database 230, and the time variation D _all (T) of the number of documents is calculated. At this time, in order to reduce processing, D _all (T) may be calculated only for the number of documents in the latest N periods.

Step 1120) An unprocessed word / phrase w is selected from the word / phrase list created in Step 1100 and stored in the buffer, the number of uses of the word / phrase for each counting period is read from the word / phrase database 230, and the frequency of use of the word / phrase for each counting period is read. A function D _w (T) representing the variation of is obtained. At this time, D _w (T) may be calculated only for the number of documents used in the most recent N period in order to reduce processing.

Step 1130) The similarity R (w) is obtained by evaluating the similarity between the function D _w (T) representing the variation in the usage frequency of the word w and the function D _all (T) representing the temporal variation in the number of documents. The similarity is calculated using the following equation which is a general waveform correlation function.

ステップ１１４０） 波形の相関度Ｒ（ｗ）を元に、補正された語句ｗの使用頻度の時間変化を表す関数Ｇ_w（Ｔ）を算出する。Ｄ_w（Ｔ）の最小値をminD_w、最大値をmaxＤ_w，Ｄ_all（Ｔ）の最小値をminＤ_all（Ｔ）、最大値をmaxＤ_all（Ｔ）とし、次式で求める。

Step 1140) Based on the correlation degree R (w) of the waveform, a function G _w (T) representing a temporal change in the usage frequency of the corrected word w is calculated. The minimum value of D _w (T) is minD _w , the maximum value is maxD _w , the minimum value of D _all (T) is minD _all (T), and the maximum value is maxD _all (T).

この際、計算の簡略化のため、Ｒ（ｗ）が０よりも小さい場合には、Ｒ（ｗ）＝０とみなして計算してもよい。また、Ｇ_w（Ｔ）が０よりも小さくなる場合には、Ｇ_w（Ｔ）＝０としてもよい。

At this time, in order to simplify the calculation, when R (w) is smaller than 0, it may be calculated assuming that R (w) = 0. When G _w (T) is smaller than 0, G _w (T) = 0 may be set.

Step 1150) The time change G _w (T) of the corrected usage frequency of the obtained phrase w is recorded in a phrase frequency buffer (not shown) of the topic level calculation unit 1050.

  Step 1160) It is confirmed whether or not an unprocessed word / phrase exists in the word / phrase list, and if it exists, the process returns to Step 1120 to continue the processing. If there is no unprocessed word, the process ends.

  Next, a processing flow of the topic level calculation unit 260 is shown.

  FIG. 12 is a flowchart of processing of the topic level calculation unit in the second embodiment of the present invention.

Step 1210) topics calculation unit 1050 starts the processing for each predetermined period in the same manner as the phrase frequency calculating unit 1040, acquires the time t _p of the start of processing.

Step 1220) Next, based on the processing target time corresponds to the time range from _{t p} obtained using the positive value _{t q} previously given to _t p -t _q, topic weighting function It _p (t ). At this time, it is preferable to create a weighting function such as the impact curve shown in Patent Document 1 described above.

Step 1230) Take out one G _w (T) representing a temporal change in the frequency of use of a certain word w from a word buffer (not shown) and multiply it by the topic weighting function It _p (t) to obtain the topic of the word w Determine the degree TS (w).

ステップ１２４０） 得られた語句ｗの話題度ＴＳ（ｗ）と語句ｗと処理時間ｔ_ｐとを話題度記録装置１０６０に出力する。例えば、
「決勝戦 32.8 2006/016 13:30」
といった結果が出力される。一連の処理で出力される話題度情報は全て同じ時刻情報を持つため、記憶量の軽減のためそれらをまとめて記録してもよい。話題度記録装置１０６０に出力される話題度情報の例を図１３に示す。

Step 1240) topics degree TS of the resulting word w and (w) outputs the word w and the treatment time _{t p} the topic of the recording apparatus 1060. For example,
"Final 32.8 2006/016 13:30"
Will be output. Since all topic level information output in a series of processes has the same time information, they may be recorded together to reduce the storage amount. An example of topic level information output to the topic level recording apparatus 1060 is shown in FIG.

  Step 1250) It is confirmed whether or not the unprocessed phrase frequency information is included in the phrase frequency buffer (not shown). If included, the process returns to Step 1130 to continue the processing. If not included, the process ends.

  Further, the operation of each component of the topic level calculation device described above can be constructed as a program, installed on a computer used as the topic level calculation device and executed, or distributed via a network.

  Further, the constructed program can be stored in a portable storage medium such as a hard disk, a flexible disk, or a CD-ROM, and can be installed or distributed on a computer used as a topic level calculation device.

  The present invention is not limited to the above-described embodiment, and various modifications and applications can be made within the scope of the claims.

  The present invention can be applied to a technique for extracting a topic / phrase from a document group.

It is a figure for demonstrating the principle of this invention. It is a principle block diagram of this invention. It is a block diagram of the topic degree calculation apparatus in the 1st Embodiment of this invention. It is an example of the information accumulate | stored in the phrase database in the 1st Embodiment of this invention. It is a flowchart of a process of the phrase frequency calculation part in the 1st Embodiment of this invention. It is an example of the time fluctuation | variation of the phrase frequency in the 1st Embodiment of this invention. It is an example of the time fluctuation | variation of the number of documents in the 1st Embodiment of this invention. It is an example of the time fluctuation | variation of the phrase frequency except the influence of the fluctuation | variation of the number of documents in the 1st Embodiment of this invention. It is a flowchart of the process of the topic degree calculation part in the 1st Embodiment of this invention. It is a block diagram of the topic degree calculation apparatus in the 2nd Embodiment of this invention. It is a flowchart of a process of the phrase frequency calculation part in the 2nd Embodiment of this invention. It is a flowchart of a process of the topic degree calculation part in the 2nd Embodiment of this invention. It is an example of the information accumulate | stored in the topic degree recording apparatus in the 2nd Embodiment of this invention.

Explanation of symbols

200 Document Database 210 Document Analysis Unit, Document Analysis Unit 220 Phrase Counting Unit 230 Phrase Database 240 Phrase Frequency Calculation Unit, Phrase Frequency Calculation Unit 250 Topic Level Calculation Unit, Topic Level Calculation Unit 260 Topic Display Device 1040 Phrase Frequency Calculation Unit 1050 Topic Level Calculation unit 1060 Topic level recording device

Claims (10)

A topic level calculation method in an apparatus that analyzes a large number of documents and determines the strength of topicality at a desired time with respect to a phrase included in the document,
When an input document group having creation time information is input, the document analysis unit analyzes the input document group and extracts a word / phrase (hereinafter referred to as an in-document word / phrase) to be evaluated. input document number, the document analyzing step of aggregating the word frequency, stored in the word database along with them aggregated aggregation time and the documents in the word is the number of times of use of the document the phrase,
The phrase frequency calculation means is a phrase (hereinafter referred to as a processing target phrase w ) that is subject to topic level calculation processing input from the outside, and time information (hereinafter, calculation designation) that indicates the topic at which the topic is calculated. An input reception step for acquiring the time)
The phrase frequency calculation means searches the phrase database based on the processing target phrase, and the time variation of the phrase frequency based on the phrase frequency of the phrase in the document corresponding to the processing target phrase w and the aggregation time A phrase frequency calculating step for calculating a phrase frequency function G _w (T) representing
The topic frequency calculation means calculates the topic frequency indicating the level of topicality of the processing target phrase at the time of the input time input in the input reception step, the phrase frequency function G calculated in the phrase frequency calculation step a topic degree calculation step to obtain by multiplying _w (T) by a topic weight function for calculating a high topic degree when the phrase frequency is increasing near the calculation designated time;
And
In the phrase frequency calculating step,
The word frequency calculation means obtains a time fluctuation D _w (T) of the word frequency in the input document of the processing target word w and a time fluctuation D _all (T) of the number of input documents, and D _w ( T) and by evaluating the similarity of D _all (T) obtaining a correlation degree R (w), a value proportional to a value obtained by multiplying the R (w) to the D _all (T), said D _w A topic degree calculation method characterized in that a value subtracted from (T) is a phrase frequency function G _w (T) . In the phrase frequency calculating step,
The correlation degree R (w) between the time fluctuation D _w (T) of the word frequency in the input document group of the processing target word w and the time fluctuation D _all (T) of the number of input documents

Asking for,
The topic level calculation method according to claim 1 . In the phrase frequency calculating step,
Using the degree of correlation R (w) between the time variation D _w (T) of the word frequency in the input document of the processing target word w and the time variation D _all (T) of the number of input documents, the number of input documents The phrase frequency function G _w (T) considering the increase or decrease of the phrase frequency due to the fluctuation of

Asking for,
The topic level calculation method according to claim 1 . In the input receiving step,
Instead of accepting the processing target phrase w that is the subject of the topic level calculation process and the calculation designated time indicating the topic level at which the topic level is calculated,
In the phrase frequency calculating step,
Obtaining a phrase frequency function G _w (T) for all the processing target phrases included in the input document group;
In the topic level calculation step,
Obtain the topic level at the processing time for all the processing target words included in the input document group,
Topical calculation method of claims 1 to 3, wherein. A topic degree calculation device that analyzes a large number of documents and determines the strength of topicality at a desired time with respect to a phrase included in the document,
When an input document group having creation time information is input, the input document group is analyzed to extract a word / phrase (hereinafter referred to as an in-document word / phrase) for topicality evaluation, and the number of input documents input, Document analysis means for totalizing phrase frequencies, which are the number of times the phrase is used in the document, and storing the total frequency in the phrase database together with the total time and the phrase in the document
A word / phrase (hereinafter, referred to as a processing target phrase / w ) input from an externally inputted topic degree calculation process is acquired, the word / phrase database is searched based on the processing target phrase / phrase, and the acquired processing target phrase / w A phrase frequency calculating means for calculating a phrase frequency function G _w (T) representing a temporal variation of the phrase frequency based on the phrase frequency of the corresponding phrase in the document and the counting time;
Obtain time information (hereinafter referred to as a calculation designated time) indicating the time at which the topic is calculated from the outside, and at the time of the calculation designated time, obtain a topic level representing the level of topicality of the processing target phrase. , Multiplying the phrase frequency function G _w (T) calculated by the phrase frequency calculation means by a topic weight function for calculating a high topic level when the phrase frequency increases near the input time. The topic level calculation means obtained by
Have
The phrase frequency calculating means includes:
A time variation D _w (T) of the phrase frequency in the input document of the processing target phrase w and a time variation D _all (T) of the number of input documents are obtained, and D _w (T) and D _all (T ) To obtain a correlation degree R (w), and a value obtained by subtracting a value proportional to a value obtained by multiplying the D _all (T) by the R (w ) from the D _w (T) Including means for the phrase frequency function G _w (T) ,
A topic degree calculation device characterized by that. The phrase frequency calculating means includes:
The correlation degree R (w) between the time fluctuation D _w (T) of the word frequency in the input document group of the processing target word w and the time fluctuation D _all (T) of the number of input documents

Asking for,
The topic level calculation device according to claim 5 . The phrase frequency calculating means includes:
Using the degree of correlation R (w) between the time variation D _w (T) of the word frequency in the input document of the processing target word w and the time variation D _all (T) of the number of input documents, the number of input documents The phrase frequency function G _w (T) considering the increase or decrease of the phrase frequency due to the fluctuation of

Asking for,
The topic level calculation device according to claim 5 . The phrase frequency calculating means includes:
Instead of accepting the processing target phrase w that is the subject of the topic level calculation processing and the calculation designated time that indicates when the topic level is calculated, all the processing target phrases included in the input document group Means for determining a phrase frequency function G _w (T) for
The topic level calculation means includes:
Including means for determining a topic level at a processing time for all processing target words included in the input document group,
The topic degree calculation device according to claim 5 . On the computer,
9. A topic level calculation program that causes each means of the topic level calculation device according to claim 5 to be executed. On the computer,
9. A computer-readable recording medium storing a topic level calculation program for executing each means of the topic level calculation device according to claim 5 .

Images