JP4496900B2 - Event information extraction apparatus and program - Google Patents

Description

The present invention relates to event information extracting device及 beauty program, such as the Internet seasonal events, sporting events, in the art of providing a topic of events initiated at a specific date and time, the document information described for various events acquired from one or more information providing servers connected to the network etc., it extracts the phrases become the subject according to the attention level of the event, on the event information extracting device及 beauty program for providing.

  In recent years, with the development of computer networks such as the Internet, a large amount of electronic document information has been accumulated one after another. In particular, it has become easier for individuals to use their bulletin boards and blog services to express their opinions and opinions on matters that interest them. Therefore, if a large amount of information transmitted one after another, such as news information, bulletin boards, and blogs, is collected and analyzed, it becomes possible to grasp the latest news and events.

  Conventionally, an information flow search method has been proposed in which information transmitted from a plurality of information providing servers is classified into categories for each topic, and the time transition of the topic is presented and searched (for example, see Patent Document 1).

  In this information flow search method, processing is performed in which similar documents are collected from a document set transmitted in a certain period and assigned to a category. For this reason, since a category is assigned only after a plurality of similar documents are transmitted, the topic can be presented only after a lapse of time since the information on the topic is transmitted for the first time.

Also, personal interests include events such as events that start on a specific date and time, and most of them experience and experience their experiences and opinions. Therefore, the topic can be presented only after the event ends.
JP 2000-242652 A

  However, as described above, according to the conventional technology, a topic can be provided only after an event has already ended, and the latest event information that may become a topic cannot be provided quickly.

The present invention has been made in view of the above points, and is an event information extractor that can provide a user with an interesting phrase before an event is started for an event that starts at a specific time. and to provide a device及 beauty program.

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

The present invention relates to an event information extraction method for extracting topics related to an event that starts at a specific date and time, such as a season, an event, or a sports event.
The event name and the period of the event are acquired from the event information storage means for storing the event information set in which the event date and time and the outline are described (step 1), and the event name is acquired from the document information storage means for storing the document set having time information. And a document extracting step by event, which is stored in the event-specific document information storage means by event (step 2),
An event attention level that calculates a document that has time information within the specified counting period from the extracted document for each event , and calculates the total value of the documents up to now divided by the total number of days as the attention level Calculating step (step 3);
When an event with a high degree of attention is selected that satisfies a predetermined condition and is before the start of the event, the summary sentence of the event is morphologically analyzed at time interval T1, and the phrase is determined based on the appearance position and the number of characters of the phrase. If the document is extracted and exits after the start of the event, the document stored in the event-specific document information storage means is morphologically analyzed at a time interval T2 smaller than T1, and the number of documents appearing for each word is obtained. A phrase extraction step (step 4) is performed in which phrases with a large number of appearing documents are extracted and output to the phrase storage means.

The present invention provides an event attention level calculation step (step 3).
When the same user enters the same keyword more than once in a short time interval using information that identifies the user who entered the search term from the search terms that match the event name and requested during the specified aggregation period by counting a one-time, aggregated search terms using the number of which related to the event, regularly seeking the total value obtained by adding the aggregated value of the search words and aggregated value from the document to a regular The value obtained by dividing the total sum obtained up to the present by the total number of days is calculated as the attention level.

The present invention provides an event attention level calculation step (step 3).
For the event before the event, the past attention level that is the sum of the total values from the past documents in the past event with the same event name is searched from the end event accumulation means, and the past attention level of the searched past event is determined , Add to the sum of the document totals up to now, and calculate the value of attention divided by the number of days of past event days and the total number of days, and for events after the end, after a preset period, A value obtained by adding all the total values from documents in the event held this time is accumulated in the end event accumulation means as a past attention degree.

The present invention provides an event attention level calculation step (step 3).
For events prior to the event, search the past event accumulation means for past attention, which is the sum of the total values from past documents and the total values from the search terms for past events with the same event name. After adding the past attention degree of the past event that has been obtained to the sum of the total values obtained up to the present time, the value obtained by dividing the past event holding days and the total number of days is calculated as the attention degree. For the event, a value obtained by adding all the total values in the event held at this time is stored in the end event storage means as a past attention level after a preset period.

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

The present invention (Claim 1) is an event information extraction device 100 that extracts topics related to events that start at a specific date and time, such as seasons, events, and sports events.
The event name and holding period are acquired from the event information storage unit 201 that stores the event information set that describes the event date and time and the summary, and the event name is included from the document information storage unit 202 that stores the document set having time information. An event-specific document extraction unit 101 that searches and extracts documents and stores them in the event-specific document information storage unit 103 by event;
An event attention level that calculates a document that has time information within the specified counting period from the extracted document for each event , and calculates the total value of the documents up to now divided by the total number of days as the attention level Calculating means 102;
When an event with a high degree of attention is selected that satisfies a predetermined condition and is before the start of the event, the summary sentence of the event is morphologically analyzed at time interval T1, and the phrase is determined based on the appearance position and the number of characters of the phrase. If it is extracted and after the start of the event, the document stored in the event-specific document information storage means 103 is morphologically analyzed at a time interval T2 smaller than T1, and the number of documents appearing for each phrase is determined. And a phrase extracting unit 104 that extracts a phrase having a large number of found documents and outputs the extracted phrase to the phrase storing unit 204.

According to the present invention (claim 2), the event attention level calculation means 102
When the same user enters the same keyword more than once in a short time interval using information that identifies the user who entered the search term from the search terms that match the event name and requested during the specified aggregation period by counting a one-time, aggregated search terms using the number of which related to the event, regularly seeking the total value obtained by adding the aggregated value of the search words and aggregated value from the document to, Means for calculating as a degree of attention a value obtained by dividing the total sum obtained up to now by the total number of days .

The present invention (Claim 3) further includes an end event accumulation unit that stores past attention levels corresponding to past event names,
In event attention level calculation means 102,
For events before the event, search the past attention level from the end event storage means, which is the sum of the aggregated values from past documents for past events with the same event name , added to the sum of the aggregate value of the document up to, to calculate the value obtained by dividing the number of days for which the sum of the aggregate number of days in the past of the event held on the number of days as the degree of attention, for after the end of the event, after a period which is set in advance, this time Means for accumulating in the end event accumulating means as a past attention level a value obtained by adding all the total values from documents in the held event .

The present invention (Claim 4) further includes an end event accumulation means in which past attentions corresponding to past event names are stored,
In event attention level calculation means 102,
For the event before the event, the past event degree is searched from the end event accumulation means, which is the sum of the total value of the total value from the past document and the total value from the search word in the past event having the same event name, The past attention degree of the past events searched is added to the total sum obtained up to the present , and the value obtained by dividing the past event holding days and the total number of days is calculated as the attention degree. The event after the end includes means for accumulating a value obtained by adding all the total values of the currently held event as a past attention degree in the end event accumulating means after a preset period.

The present invention (Claim 5 ) is an event information extraction program for causing a computer to function as each means of the event information extraction apparatus according to any one of Claims 1 to 4 .

  As described above, according to the present invention, the degree of attention of an event held for a specific period is calculated, and for an event to be started from now on, event information that introduces the content (for example, a guide text of the organizer) is included. By extracting a phrase from, information that may become a topic before the event is started can be provided.

  In addition, after the event is started, it is possible to extract and provide a word / phrase that is a topic from document information (for example, word-of-mouth information) transmitted by people watching the event.

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

  FIG. 3 shows the configuration of the event information extraction device in one embodiment of the present invention.

  The event information extraction apparatus 100 shown in the figure receives information from the event information storage unit 201, the document information storage unit 202, and the search word storage unit 203 and outputs the extracted words to the word storage unit 204.

  The event information extraction device 100 includes an event information extraction unit 101, an event attention level calculation unit 102, an event-specific document information storage unit 103, a phrase extraction unit 104, and an end event storage unit 105.

  The event information storage unit 201 stores event information. FIG. 4 shows an example of the event information storage unit 201. As shown in the figure, the event information includes information related to events such as an event name 401, a start date 402, an end date 403, a time 404, a place 405, and a summary sentence 406.

  The document information storage unit 202 collects newly generated document information and updated document information from document information published on the Internet such as a home page, blog, and bulletin board, and creates a time such as a creation date and an update date. Accumulated with information. From these document information, it is possible to collect impressions of people who have experienced the event and news articles that introduce the event status.

  The search word storage unit 203 collects search words input by the user at a search site that provides a search service on the Internet and stores them together with the time when the search is requested. Search terms are words that are input as a clue to search for information that the user is interested in, so by collecting and analyzing the search terms regularly, you can know the information that the user is highly interested in. it can.

  The event attention level calculation unit 102 periodically acquires event information from the event information storage unit 201 and manages the event information by event name. Even if the event information is short, it is considered that it is registered in the event information storage unit 201 a few days before the event, so even if the event attention level calculation unit 102 executes the process once a day, the event attention level before the event is held. It is possible to calculate.

  Also, when the event ends, new information about the event cannot be collected (the impressions and experiences of people who participated in the event tend to disappear within a few days after the event ends), and are deleted from the managed event.

  FIG. 5 shows an example of information accumulated in the event attention level calculation unit 102. FIG. 6A shows an example of the structure of the event totaling result table 500. New event information is acquired from the event information storage unit 201, and an event ID 501 that is a unique number for identifying the event is assigned to each event. Event information is managed.

  The event name 502 is the name of the event, the start date 503 is the start date of the event, the end date 504 is the end date of the event, and the time 505 records the start time and end time at which the event is held during the day. Is done.

  The past attention level 506 is the attention level when a similar event has been held in the past, and is recorded by searching the end event accumulation unit 105 when adding a new event to the event-by-event totaling result table 500.

  The next word / phrase extraction date / time 507 is time information for controlling the interval for extracting the word / phrase related to the event, and the word / phrase extraction date / time is updated when the word / phrase extraction unit 104 performs the word / phrase extraction process.

  The count end date and time 508 is set to the date and time after the event ends, and events that have passed the count end date and time 508 are deleted from the event-based count result table 500.

  The total value 509 is data used when calculating the degree of attention of an event, and data for a certain period can be accumulated for a specific period. Further, a ranking of event attention level as shown in FIG. 5B is generated from the value of the total data 509 and the value of the past attention level 506. Even if the event attention level is calculated at a very short interval, there is no change. On the other hand, if the event attention level is calculated at a very long interval, there is a problem that the time elapses after the attention starts. Further, when a new event is added, it is necessary to calculate the degree of attention of the added event. Therefore, after a new event is added to the event totaling result table 500, the event attention ranking ranking generation process is executed. For example, if the event addition process is once a day, the ranking generation process is also executed once a day.

The event information extraction unit 101 is executed independently of the processing in the event attention level calculation unit 102. New search words are successively stored in the search word storage unit 201, and new document information collected one after another is stored in the document information storage unit 202 in the same manner. The event information extraction unit 101 periodically acquires information stored in the search word storage unit 201 and the document information storage unit 202, and extracts only information related to the event. The extraction interval here is the minimum interval for providing information to the user. First, an event name is acquired from the event attention level calculation unit 102, and search words that match the event name are tabulated from the search words acquired from the search word storage unit 203. Further, document information including event names is totaled for each event name from the document information acquired from the document information storage unit 202. Then, the total value aggregated from the search term and the total value aggregated from the document information are added and recorded in the total value column 509 of the event-specific total result table 500 of the event attention level calculation unit 102. Also, document information related to the event is organized for each event and recorded in the event-specific document information storage unit 103.

  After the processing of the event information extraction unit 101 ends, the word / phrase extraction unit 104 acquires an event name with a high degree of attention from the event attention level calculation unit 102. A summary sentence is acquired, a topic phrase is extracted and output to the phrase storage unit 204. After an event with a high degree of attention is held, document information is acquired from the event-specific document information storage unit 103, a word representing a topic is extracted, and is output to the word storage unit 204. The completed event is not acquired as an event with a high degree of attention after several days, and the extraction of words is also completed.

  Even if there is document information written by people who are interested in the event before the event is held, the information does not contain more information than the summary text of the event. Extract words. In addition, since the event information in the event information storage unit 201 is not frequently updated (once it is registered, it is updated only when there is a change), so the word / phrase extraction process before the event is frequently performed. There is no need. For example, once a day is considered sufficient. More efficiently, identification information for identifying event information registered and updated in the event information storage unit 201 is assigned to each event, and the phrase extraction process is executed only during registration and update by checking the identification information. You may comprise. In this case, it is necessary to reset the above-described identification information when the phrase extraction process is completed.

  During the event period, people interested in the event release new document information one after another, so the interval of the phrase extraction process is shortened and the latest information can be provided to users one after another. After the event, new information is reduced, and users can only refer to the next time to provide topics about the event after the event. Therefore, the word / phrase extraction process after the end of the event is set longer than during the event holding period. In this way, it is possible to frequently execute the phrase extraction process during the event and always provide the latest topics, and by performing the phrase extraction process before and after the event at intervals, It is possible to reduce the load of the phrase extraction process.

  FIG. 6 shows an example of storage of words output to the word storage unit 204. The extracted word / phrase 1004 is stored in association with the event name 1001, the event start date 1002, and the event end date 1003. By periodically executing the processing of the event information extraction unit 101 and the phrase extraction unit 104, an event with a high degree of attention and a phrase representing the topic of the event are output one after another. The phrase storage unit 204 records the start date 1002 and the end date 1003 of the event, so what is the subject of the attention event currently being held or what attention event will be held in the future? Such information can be provided.

  In FIG. 7, the timing of each process of the event information extraction apparatus 100 is illustrated with the event name “event A” as an example. First, “event A” is registered in the event information storage unit 201 (FIG. 4, 401). When the post-registration event attention calculation unit 102 is activated, “event A” is added to the event totaling result table 500 of FIG. 5 and the event ID “0000101” is assigned and managed. The start date 402, the end date 403, and the time 404 are copied to the start date 503, the end date 504, and the time 505 in FIG.

  Subsequently, the end event accumulation unit 105 is searched for “event A”, and it is checked how much attention has been given to the similar event in the past. As shown in FIG. 8, the end event accumulation unit 105 includes an event name 1101, a start date 1102, an end date 1103, and a past attention level 1104. The event name of the end event accumulation unit 105 is searched with the event name “event A”. At this time, if “event A” includes a year or the like, a search is performed using a character string excluding the year, and an event name including the character string is obtained from the event name 1101 column of the end event accumulation unit 105. To do. If a plurality of event names can be acquired as a result of the search, a new event with an end date is selected, and the corresponding past attention level 1104 is acquired. The example of the event totaling result table 500 in FIG. 5 is an example in which an event could not be acquired as a search result of “event A”, and the past attention level is blank. Further, “event C” is an example in which “event X” is acquired as a result of searching the past event accumulation unit 105, the past attention level “30” is acquired, and the past attention level 506 in the event-by-event total result table 500 is acquired. “30” is recorded.

  The column for the next word extraction date and time 507 when a new event is added is left blank.

  The aggregation end date and time 508 is set to “September 5, 2004, 17:00: 00” five days after the event end date and time. Events that have passed the aggregation end date and time are deleted from the event-specific aggregation result table 500. The totaling end date and time 508 is set by adding several days to the event end date and time, and the number of days to be added is determined in advance by examining the transition of the number of documents and search terms related to the event. The number of days to be added may be set to the same number of days regardless of the event, or the number of days may be set for each event type in consideration of the case where the speed at which people's interest disappears differs depending on the event type. For example, in the case of an event that takes place only for one day such as a fireworks display, people tend to lose interest in about two days. When setting the number of days for each type, a rule for determining the type (for example, including “fireworks” in the event name) is defined, and the number of days determined for each type is set for an event that matches the rule.

  Subsequently, the event attention level calculation unit 102 generates an event attention level ranking and records it in the event attention level table 510.

  FIG. 9 is a diagram showing a flow of ranking generation processing of the event attention level calculation unit according to the embodiment of the present invention.

  First, all event ID sets are acquired from the event totaling result table 500 (step 101), one event is extracted from the event ID set (step 103), and it is checked whether the event is before the event (step 104). If it is before the event (step 104, Yes), the event attention level is calculated by the following formula based on the past attention level 506 and the total value up to the present time (step 105).

イベントが開催されている場合（ステップ１０４、Ｎｏ）、集計終了日時５０８を過ぎているか調べ（ステップ１０６）、集計終了日時５０８を過ぎていない場合は（ステップ１０６、Ｎｏ）、集計値５０９の現在までの値を全て加算し、イベント開催日から現在日までの日数で除算した値をイベント注目度として算出する（ステップ１０７）。また、集計終了日時を過ぎている場合は（ステップ１０６、Ｙｅｓ）、集計終了日時までの集計値を全て加算した値を算出し、終了イベント蓄積部１０５の過去注目度１１０４の欄に記録する（ステップ１０８）。続いて、イベント別集計結果テーブル５００からイベントを削除する（ステップ１０９）。

When the event is held (step 104, No), it is checked whether or not the counting end date and time 508 has passed (step 106). When the counting end date and time 508 has not been passed (step 106, No), the current value of the counting value 509 is determined. All the values up to are added, and a value obtained by dividing by the number of days from the event date to the current date is calculated as the event attention degree (step 107). If the total end date / time has passed (step 106, Yes), a value obtained by adding all the total values up to the total end date / time is calculated and recorded in the past attention degree 1104 column of the end event storage unit 105 ( Step 108). Subsequently, the event is deleted from the event totaling result table 500 (step 109).

  When processing that matches the conditions from step 103 to step 109 is executed for all event ID sets acquired in step 101 and the processing is completed for all event IDs (step 102, Yes), for each event ID The event attention degrees calculated in the above are ranked in descending order of the attention degree and output to the event attention degree table 510. In this example, the current event attention level is the average number of attentions per day, and the past attention level is the number of attentions over the entire period in which people were interested before and after the event was held. For events that are always of high interest, the average number of attention per day (= event attention) will be approximately constant, and for events that are of high interest only immediately after the event has started, the event duration will increase. The average number of attentions per day (= event attention level) decreases.

The event information extraction unit 101 always acquires an event name from the event attention level calculation unit 102 at regular intervals, and extracts information related to the event. Therefore, after “event A” is added to the event totalization result table 500 of the event attention level calculation unit 102, when the processing of the event information extraction unit 101 is started, document information related to “event A” starts to be extracted. . The event information extraction unit 101 executes processing periodically (for example, at an interval of one hour) for a period until “event A” is deleted from the event totaling result table 500. By the processing of the event information extraction unit 101, the total value of the document information and the search word is recorded every hour in the total value 509 of the event-specific total result table 500. Also, the document information organized by event is accumulated in the document information accumulation unit 103 by event.

  The process flow of the event information extraction unit 101 will be described with reference to FIG.

  FIG. 10 is a diagram showing a flow of processing of the event information extraction unit in one embodiment of the present invention.

  The event information extraction unit 101 first acquires an event ID and an event name from the event attention level calculation unit 102 (step 201). Next, a search term is acquired from the search term storage unit 201 by specifying a time range (step 202), and search terms that match the event name are tabulated for each event name (step 203). The time range specified here is one hour before the current time if processing is executed every other hour, and the time range specified next time is one hour consecutive to the time range specified this time. To do.

  It is conceivable that the same user inputs the search term multiple times. Therefore, if the same user inputs the same keyword a plurality of times at short time intervals based on information (for example, cookie information of the Web browser) that identifies the user who has input the search word, the search is performed by counting once. Aggregate the number of people entering each word.

  Next, the document information storage unit 202 also points out the time range and acquires the document information (step 204), selects the document information including the event name as the document information related to the event, and counts the number of documents for each event name. To do. The time range specified here is the same time range as the time range specified in step 202. Document information related to the event is organized for each event and recorded in the event-specific document information storage unit 103 (step 205).

  When selecting a search term or document information related to an event, the event name was used. However, if the event name includes a year or place name (for example, XX in Tokyo), enter the year or place name. Using the excluded character string, a related search term and document information are selected.

  In addition, as a storage method of the event-specific document information storage unit 103, an area for storing document information is set for each event ID, and the file name including the time is included in the document information so that the time information can be associated with the area. Accumulate by adding, for example.

  Then, for each event ID, the total value of the search terms and the total number of documents calculated from the document information are added and recorded in the total value 509 of the event-specific total result table 500 of the event attention level calculation unit 102 (step 206). . The total value of the search term is the number of users who have made a search request with the event name, and can be said to be the number of people interested in the event. In addition, the total number of documents represents the number of people who are interested in the event because the comments and opinions about the event are documented. That is, the total data 509 stores the number of people interested in the event for each period.

  Similarly to the event information extraction unit 101, the phrase extraction unit 104 always executes processing at regular intervals (for example, one hour interval). N event IDs with a high degree of attention are obtained from the event attention level table 510 of the event attention level calculation unit 102, and a phrase (topic word) representing the topic of the event is extracted. For this reason, when “event A” is listed as N events with high attention level in the event attention level table 510 and the processing of the phrase extraction unit 104 is started, topic words related to “event A” begin to be extracted. . If “Event A” is not selected as an N event with a high degree of attention, or if the event is deleted from the event-by-event aggregation result table 500 after the aggregation end date and time, the topic word for “Event A” is Not extracted.

  The phrase extraction unit 104 extracts topic words for N events having a high degree of attention each time it is executed, and outputs the topic words to the phrase storage unit 204.

  The flow of the phrase extraction unit 104 will be described with reference to FIG.

  FIG. 11 is a diagram showing the flow of the phrase extraction unit in one embodiment of the present invention.

  The phrase extraction unit 104 is activated after the event information extraction unit 101 finishes processing. That is, if the process of the event information extraction unit 101 is every other hour, the process of the phrase extraction unit 104 is also executed every hour.

  First, the top N event IDs 511 having a high level of attention are acquired from the event level of attention table 510 of the event level of interest calculation unit 102 (step 301). Next, one event ID is extracted, and the event name 502, start date 503, end date 504, time 505, next word extraction date / time 507, count end date / time corresponding to the event ID having the high degree of attention are collected from the event totaling result table 500 508 is acquired (step 303). Check whether the next word extraction date / time is empty (processed for the first time with a registered event) or whether the current date / time has already passed the next word extraction date / time (step 304). Whether or not (step 305). If it is before the event, add the time T1 to the current time to set the next word extraction date and time (step 306), obtain a summary sentence from the event information storage unit 201 (step 307), extract the word The event name, start date, and end date are output to the phrase storage unit 204 (step 308). Since the time T1 to be added here is before the event is held, a somewhat long time is sufficient, for example, one day (24 hours). Also, since it is before the event, even if there is document information written by people who are interested in the event, it does not contain detailed content beyond the outline of the event, so the event information storage unit 201 A phrase is extracted from the summary sentence 406.

  Next, when it is not before the event (step 305, No), it is checked whether the event is being held (step 309), and when it is being held (step 309, Yes), the time at the start time of the phrase extraction process T2 is added to set the next word extraction date (step 310). During the event period, document information described by people interested in the event can be collected one after another. Therefore, a document information set related to the event is acquired from the event-specific document information storage unit 103 (step 311), and the phrase is extracted from the document information set. Are extracted and output together with the event name, start date, and end date to the phrase storage unit 294 (step 312). Since the time T2 to be added in step 310 is during the event holding period, a short time is set. Here, for example, one hour is set in accordance with the processing interval of the event information extraction unit 101. Even if it is executed in a shorter time than the processing interval of the event information extraction unit 101, new topic information cannot be extracted because new document information is not stored in the event-specific document information storage unit 103. That is, if it is desired to execute the phrase extraction in a shorter time, the processing interval of the event information extraction unit 101 needs to be further shortened.

  Next, when the event is not being held (step 309, No), the event has already ended, and the time / phrase extraction date / time is set by adding the time T3 to the start time of the word / phrase extraction process (step 309). 313). Since this time T3 is after the event has ended, it is not necessary to set a short time, and is, for example, 12 hours. Since there is a possibility that new document information can be acquired even after the event ends, a document information set is acquired from the event-specific document information storage unit 103 even after the event ends (step 311), and words are extracted (step). 312).

  The process is executed for all N event IDs (step 302, Yes), and the process of the phrase extraction unit 104 is terminated.

  In the above embodiment, the phrase is extracted regardless of the difference in the attention level of the event. However, the higher the degree of attention, the shorter the word extraction interval, and the latest information can be provided at a shorter interval. You may comprise as follows.

  In addition, for events such as sports and fights where new information is generated each time it is held, the latest information can be provided by setting a short word extraction interval during the holding period. On the other hand, for events such as exhibitions that do not change much in the content of the event itself, it is not a problem to set the word extraction interval as long as about once a day even during the holding period. In this way, the phrase extraction interval may be changed according to the event type.

  Furthermore, even during the holding period, the phrase extraction interval may be set to be short only for a specific period including the time when the event is performed, based on the time information when the event is actually performed. .

  Next, the word / phrase extraction method in step 308 will be described in more detail.

  In step 308, the word / phrase extraction unit 104 receives the summary sentence from the event information storage unit 201 acquired in step 307. First, a morphological analysis process is performed on the summary sentence, and it is decomposed into morphemes such as individual parts of speech and punctuation marks. From the decomposed morphemes, nouns, compound nouns in which multiple nouns are continuous, and noun phrases that function in the same way as nouns are extracted for all consecutive words, such as “Koizumi administration support rate”. This is because nouns, compound nouns, noun phrases and the like are more suitable as verbal words and adjectives as words representing topics. In the following description, nouns, compound nouns, and noun phrases are collectively referred to as words. Since the summary sentence has the property that important information to be conveyed at the beginning of the sentence is described, a higher evaluation value A is given to the extracted word / phrase as it appears at the beginning. Moreover, since more accurate information can be conveyed as the number of characters increases, a higher evaluation value B is given as the number of characters increases. The word with the highest evaluation value is extracted from the evaluation values A and B.

  FIG. 6 is an example of data output by the phrase extraction unit 104 that started processing at “2004/07/20 21:00”. The event name “event C” in the figure before the event is held, and “the airborne Niagara” and “fireworks praise” stored in association with “event C” are the words and phrases extracted in step 308 It is an example.

  Next, the phrase extraction method in step 312 will be described in more detail. In step 312, the word / phrase extraction unit 104 receives the document information set acquired from the event-specific information storage unit 103 in step 311. Morphological analysis processing is performed on each document content of the document information set to extract nouns, compound nouns, and noun phrases (= phrases). After extracting words from all document information, the number of documents appearing for each word is totaled. Here, the words included in the title are excluded from aggregation. Of the remaining phrases, M phrases that have appeared in more document information are extracted as phrases representing the topic. In FIG. 6, the event name “event A” is being extracted in step 312 such as the words “feather dinosaur” and “first 1000 people” that are stored in association with “event A” during the event holding period. This is an example of a phrase.

  Note that the operations of the event information extraction unit 101, the event attention degree calculation unit 102, and the phrase extraction unit 104 of the event information extraction device 100 in the above embodiment are constructed as programs and installed in a computer used as the event information extraction device. And can be distributed through a network.

  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 is applicable to a technique for extracting a topic related to an event that starts on a specific date and time, such as a seasonal event or a sports event, on a network.

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 information provision apparatus in one embodiment of this invention. It is a figure which shows the example of the event information storage part in one embodiment of this invention. It is a figure which shows the example of data of the event attention degree calculation part in one embodiment of this invention. It is a figure which shows the example of accumulation | storage of the phrase storage part in one embodiment of this invention. It is a figure which shows the process timing in one embodiment of this invention. It is a figure which shows the example of the completion | finish event storage part in one embodiment of this invention. It is a figure which shows the flow of the ranking production | generation process of the event attention degree calculation part in one embodiment of this invention. It is a figure which shows the flow of a process of the event information extraction part in one embodiment of this invention. It is a figure which shows the flow of a process of the phrase extraction part in one embodiment of this invention.

Explanation of symbols

100 Event Information Extraction Device 101 Event-Specific Document Extraction Unit, Event Information Extraction Unit 102 Event Attention Level Calculation Unit, Event Attention Level Calculation Unit 103 Event-Specific Document Information Storage Unit, Event-Specific Document Information Storage Unit 104 Phrase Extraction Unit, Phrase Extraction Unit 105 End event storage unit 201 Event information storage unit, event information storage unit 202 Document information storage unit, document information storage unit 203 Search word storage unit 204 Phrase storage unit, phrase storage unit 401 Event name 402 Start date 403 End date 404 Time 405 Location 406 Summary 500 Total result table 501 by event Event ID
502 Event name 503 Start date 504 End date 505 Time 506 Past attention level 507 Next word extraction date and time 508 Total end date and time 509 Total value 510 Event attention level table 511 Rank 512 Event ID
513 Attention level 1001 Event name 1002 Start date 1003 End date 1004 Phrase 1101 Event name 1102 Start date 1103 End date 1104 Past attention level

Claims (5)

An event information extraction device that extracts topics related to events that start at a specific date and time, such as seasons, events, and sports events,
The event name and duration are acquired from the event information storage means that stores the event information set that describes the date and time of the event, and the document including the event name is stored from the document information storage means that stores the document set having time information. An event-specific document extraction unit that searches and extracts and stores the event-specific document information storage unit by event,
Event attention that calculates a document that has time information within the specified aggregation period from the extracted document periodically for each event, and calculates a value obtained by dividing the total of the total values of the document up to the present by the total number of days Degree calculation means;
When an event with a high degree of attention is selected that satisfies a predetermined condition and is before the start of the event, the summary sentence of the event is morphologically analyzed at time interval T1, and the phrase is determined based on the appearance position and the number of characters of the phrase. If it is extracted and after the start of the event, the document stored in the event-specific document information storage means is morphologically analyzed at a time interval T2 smaller than T1, and the number of documents appearing for each word is calculated. A phrase extraction unit that extracts a phrase having a large number of documents that appear and outputs the phrase to the phrase storage unit;
An event information extracting device characterized by comprising: The event attention degree calculating means includes:
A search request is made within the specified counting period, and the same user inputs the same keyword a plurality of times at short time intervals using information that identifies the user who entered the search term from the search terms that match the event name. by counting and once in the case, aggregate search terms using the number of which related to the event, regularly seeking the total value obtained by adding the aggregated value of the search words and aggregated value from the document The event information extraction device according to claim 1, further comprising means for calculating, as a degree of attention, a value obtained by dividing a total sum obtained up to the present time by the total number of days . It further has an end event storage means in which past attention levels corresponding to past event names are stored,
The event attention degree calculating means includes:
For the event before the event, the past attention level that is the sum of the total values from the past documents in the past event with the same event name is searched from the end event accumulation means, and the past attention level of the searched past event is determined , Add to the total of the total values of the documents up to the present, and calculate the value obtained by dividing the number of days of past event held by the number of days added as the degree of attention. 2. The event information extracting device according to claim 1, further comprising means for accumulating in the end event accumulating means as a past attention level a value obtained by adding all the total values from the documents in the event held this time . It further has an end event storage means in which past attention levels corresponding to past event names are stored,
The event attention degree calculating means includes:
For the event before the event, the past event degree is searched from the end event accumulation means, which is the sum of the total value of the total value from the past document and the total value from the search word in the past event having the same event name, The past attention degree of the past events searched is added to the total sum obtained up to the present , and the value obtained by dividing the past event holding days and the total number of days is calculated as the attention degree. The event information extraction device according to claim 2, further comprising means for accumulating a value obtained by adding all of the total values of the currently held event as a past attention degree in the end event accumulating means for the event after the completion after a preset period. .   An event information extraction program for causing a computer to function as each means of the event information extraction device according to claim 1.

Images