JP4114462B2 - Information search device and information search system - Google Patents

Description

[0001]
BACKGROUND OF THE INVENTION
The present invention relates to an information retrieval apparatus and an information retrieval system in a document management system that manages documents (electronic documents) such as electronic mail.
[0002]
[Prior art]
2. Description of the Related Art Conventionally, document management systems for managing documents such as e-mails have been provided with functions such as full-text search for text. Actions such as date and time, location, and meeting request are extracted from the document. For example, information related to "Event where the venue is XX Dome" and "Request for reply during this week" is written. It is possible to search under various conditions using the types of words in the document, such as likely documents.
[0003]
Furthermore, in order to improve user convenience, Japanese Patent Laid-Open No. 10-69472 extracts e-mail deadlines for action from among dates and times in the text, and sorts them in order of close deadlines. A function is proposed.
[0004]
Sato et al., “Automatic Digest Generation of Electronic News” (Journal of Information Processing Society of Japan, Vol.36, No.10 pp.2371-2379), from electronic news, conference dates, venues, and paper deadlines. Etc. are extracted. These are effective as techniques for extracting only predetermined information.
[0005]
On the other hand, Japanese Patent Laid-Open No. 9-269940 proposes a function of extracting information such as date and time from the text of an electronic mail and searching for mail including the date in the text using the date as a search key. In addition, locations and matters are also extracted. With these functions, you can cut out and display the date, time, place, matter, etc. in the mail text, and sort and display the mail according to whether the date and time in the mail is before or after the current time. However, for example, the search is not performed until it is specified whether the date represents an event date or a deadline date.
[0006]
[Patent Document 1]
Japanese Patent Laid-Open No. 10-69472
[Patent Document 2]
Japanese Patent Laid-Open No. 9-269940
[Non-Patent Document 1]
“Automatic digest generation of electronic news” (Journal of Information Processing Society of Japan, Vol.36, No.10 pp.2371-2379)
[0007]
[Problems to be solved by the invention]
By the way, in the actual document information search, for example, when searching for “documents including the date of the event”, or by specifying a more specific date and time, “documents including August 3, 2002 as the date of event” In addition, it is required to be searchable under various conditions, such as when searching for “documents including August 3, 2002 as date and time” without distinguishing the date of the event. . In order to be able to search under such conditions, not only a document in a specific field but also a process for specifying a word type for various documents is required.
[0008]
In addition, even if the word type is wrong due to an extraction error, it is necessary to be able to search correctly to some extent. For example, a place that should be extracted as “date” can not be searched correctly if it can only be extracted as “date and time”, but even in such a case, it is necessary to be able to search so that there is no omission. Don't be. When trying to realize these, it is impossible to search effectively by combining any of the conventional methods.
[0009]
This is because of the following problems.
[Problem 1] Sato et al.'S "automatic generation of electronic news digest" method is effective when the field is limited, such as news related to conferences. When this method is used for documents in various fields, it is necessary to prepare various rules such as a rule for extracting information on holding a meeting and a rule for extracting information on the deadline of response requests. As pre-processing, it is necessary to select whether or not the transmitted e-mail is an e-mail related to holding a conference. However, there are cases where various information is included in one e-mail, and the sorting process is not always successful.
[0010]
[Problem 2] When extracting specific information in a limited field, such as “Automatic digest generation of electronic news” by Sato et al., The extraction result is generally in the form of a table. However, in this case, information on how accurate the extraction result is lost. Therefore, it may be difficult to use the search result because it cannot be determined whether the search result can be trusted.
[0011]
[Problem 3] In Japanese Patent Laid-Open No. 9-269940, information such as date and time is extracted from the body of an e-mail, but it is not distinguished whether the date is a deadline or a date. Therefore, it is not possible to set search conditions such as searching by the date of event. On the other hand, when extracting whether the date is the deadline or the date of the event, an extraction error or the like may occur, and a search omission may occur.
[0012]
The present invention has been made in view of the above-mentioned problems of the conventional information search apparatus or information search system, and an object of the present invention is to solve the above-mentioned problems of the conventional information search apparatus or information search system. It is an object of the present invention to provide a new and improved information retrieval apparatus and information retrieval system capable of performing the above.
[0013]
[Means for Solving the Problems]
In order to solve the above problems, according to a first aspect of the present invention, an information search apparatus for searching for a document is provided. As described in claim 1, the information providing apparatus of the present invention stores a document storage unit (6) for storing a document and character strings in the document for each classification as information for searching the document. An extraction information storage unit (5), a condition input unit (7) for inputting a character string as a search condition, and a search unit (9) for searching a document from the document storage unit based on the character string input as the search condition The search unit (9) includes a candidate selection unit (10) that selects a document by loosening the search condition so that the document is searched regardless of the classification of the character string input as the search condition, A fitness level calculation unit (11) for calculating a fitness level between the selected document and a character string as a search condition is provided.
[0014]
Further, a document input unit (1) for inputting a document stored in the document storage unit, an extraction rule storage unit (3) for storing an extraction rule for extracting a character string from the input document, and an extraction rule The extraction unit (2) that extracts the character string from the document input from the document input unit by comparing with the extraction rule stored in the storage unit, and determines the classification of the character string extracted by the extraction unit, You may make it provide the document information registration part (4) which registers a character string into an extraction information storage part.
[0015]
In this case, the classification of the character strings in the extraction information storage unit (5) is, as described in claim 6, a major classification determined according to the extraction rules stored in the extraction rule storage unit (3), and You may make it consist of the small classification which subdivided the large classification. Further, as described in claim 7, the large classification is further stored in the document stored in the extraction information storage unit (5) regardless of the extraction rule stored in the extraction rule storage unit (3). You may make it have the classification for registering a noun. Even if it cannot be originally extracted as an event or place due to an incomplete extraction rule, it is possible to eliminate omissions by registering it as a noun in the word information storage unit.
[0016]
According to such an information retrieval apparatus, [Problem 1], [Problem 2], and [Problem 3] can be solved. This will be described below.
[0017]
Extraction information storage unit (5) stores character strings in documents for each category such as “event”, “action”, “location”, “date / time”, etc. By adopting a configuration having no rules, it is possible to eliminate the process of selecting in advance what field the document is in. Thus, [Problem 1] has been resolved. Further, the extraction information storage unit (5) stores the character string in the document for each classification (major classification and minor classification), so that a detailed search can be performed. For example, when searching for the date and time, by classifying whether the date and time is the “deadline date” or “the date of the event”, it is possible to perform a search by distinguishing the “date of deadline” and the “date of event”.
[0018]
  In addition, the candidate selection unit (10) selects the document by relaxing the search condition so that the document is searched regardless of the classification of the character string input as the search condition. This is to increase the number of selectable documents in consideration of extraction errors. Specifically, the search condition can be relaxed as follows.
  1:If the string entered as a search condition is a specific value, relax the search condition so that only the specific value matches and select the document.
  2:If the character string entered as a search condition is a concrete value, relax the search condition so that only the concrete value matches, and if the character string entered as a search condition is not a concrete value, only the major classification Select documents by loosening search conditions so that matches.
[0019]
In this way, by selecting a document by loosening the search condition, it is possible to prevent a search omission even if the extraction result using the extraction rule is wrong. Thus, [Problem 3] has been resolved.
[0020]
  For the document selected by the candidate selection unit as described above, the fitness level calculation unit (10) calculates the fitness level. Specifically, the fitness can be calculated as follows.
  1:Calculates the degree of matching between the selected document and the character string that is the search condition, using the classification of the character string in the selected document.
  2:Calculates the degree of matching between the selected document and the character string that is the search condition, using the positional relationship of the character string in the selected document.
[0021]
In particular, as shown in (2), when extracting information for searching a document, words representing the date, place, etc. in the document are not stored in a table format, and the respective position information is stored. At the time of retrieval, the degree of matching with the retrieval condition is calculated using the type of each word and the positional relationship between the words. By arranging and displaying to the user in descending order of relevance, the user can determine whether the search result can be trusted. Thus, [Problem 2] has been resolved.
[0022]
  In order to solve the above-mentioned problem, according to the second aspect of the present invention,,An information search system is provided that includes a server unit (100) and a terminal unit (110) connected via a network (120) and searches for a document.
[0023]
In the information search system of the present invention, the server unit (100) includes a communication unit (13) for transmitting / receiving search conditions and search results, a document storage unit (6) for storing documents, and information for searching for documents. And an extraction information storage unit (5) for storing character strings in the document for each classification, and a search unit (9) for searching the document from the document storage unit based on the character string input as a search condition. It is characterized by that.
[0024]
In the information search system of the present invention, the terminal unit (110) includes a communication unit (14) for transmitting / receiving a search condition and a search result, and a condition input unit (7) for inputting a character string as a search condition. The search unit (9) is selected with the candidate selection unit (10) that selects the document by relaxing the search condition so that the document is searched regardless of the classification of the character string input as the search condition. A fitness level calculation unit (11) that calculates the fitness level of a document and a character string as a search condition is provided.
[0025]
  According to such an information retrieval system, in addition to having substantially the same effect as the information retrieval device according to the first aspect of the present invention, for example,,The present invention can also be applied to a case where e-mails are received via a network such as the Internet, the received e-mails are sequentially registered, and searched according to a user request. Further, the function of the server unit (100) of the present invention can be provided as a so-called ASP (Application Service Provider) service, and can be developed as a business.
[0026]
DETAILED DESCRIPTION OF THE INVENTION
Exemplary embodiments of an information search apparatus and an information search system according to the present invention will be described below in detail with reference to the accompanying drawings. In the present specification and drawings, components having substantially the same functional configuration are denoted by the same reference numerals, and redundant description is omitted.
[0027]
(First embodiment)
In the present embodiment, an information processing apparatus capable of registering a date and time, a place, an event, and an action included in a document and searching using the registered information will be described. In particular, the following approaches are taken to solve the above three problems.
[0028]
For [Problem 1], a configuration that does not have an extraction rule for each field in advance. This eliminates the need to select in advance what field the document is in.
[0029]
For [Problem 2], at the time of word extraction, words representing the date, place, etc. in the document are not stored in a table format, and each position information is stored. At the time of retrieval, the degree of matching with the retrieval condition is calculated using the type of each word and the positional relationship between the words. By displaying the relevance level to the user or arranging and displaying the relevance level to the user, the user can determine how much the search result can be trusted.
[0030]
For [Problem 3], in the process of extracting the word type, it is distinguished whether the date and time is the deadline or the date of the event. In this way, for example, the user can make a search by distinguishing the event date and the deadline date. On the other hand, considering the possibility that the extraction result is wrong, the search condition set by the user to search for candidate documents is used regardless of the type of word specified by the user as a search condition so that there are no omissions. , Try to relax the search conditions.
[0031]
FIG. 1 is an explanatory diagram showing a system configuration of the information search apparatus 20 according to the present embodiment. As shown in FIG. 1, the information retrieval device 20 includes a document input unit 1, an extraction unit 2, an extraction rule storage unit 3, a document information registration unit 4, an extraction information storage unit 5, and a document storage unit 6. And a condition input unit 7, a display unit 8, and a search unit 9. The search unit 9 includes a candidate selection unit 10, a fitness calculation unit 11, and a candidate alignment unit 12.
[0032]
Among the constituent elements of the information search device 20, the document input unit 1, the extraction unit 2, the extraction rule storage unit 3, the document information registration unit 4, the extraction information storage unit 5, and the document storage unit 6 register document information. It is a component for. The condition input unit 7, the display unit 8, and the search unit 9 are components for searching for document information. Each component will be described in detail below.
[0033]
(Document input part 1)
The document input unit 1 receives a document to be registered.
[0034]
(Extractor 2)
The extraction unit 2 extracts information that matches the document input from the document input unit 1 by matching with the extraction rules stored in the extraction rule storage unit 3. Specifically, when the input document is an email, the date and time, event, location, action, etc. included in the subject or text are extracted. These dates, events, places, actions, etc. are referred to as “major classification”. In addition, for example, for an event, information that subdivides whether the event is a seminar or a meeting is extracted. These subdivided information is referred to as “small classification”.
[0035]
(Extraction rule storage unit 3)
The extraction rule storage unit 3 stores an extraction rule for extracting predetermined information from the document input from the document input unit. FIG. 2 is an explanatory diagram showing an example of the extraction rules stored in the extraction rule storage unit 3. For example, rule A-1 in FIG. 2A is an extraction rule for “event-conference”, and matches a character string such as “9th Environmental Forum”. Rule A-2 is an extraction rule in which the major classification is an event and the minor classification is not specified, and character strings such as “Gion Festival” and “Summer Festival” match. The same applies to the rules A-3 and A-4.
[0036]
Further, as shown in FIG. 2B, the extraction rule storage unit 3 also stores extraction rules relating to headwords. The extraction rule related to the headword is, for example, the headword is extracted as the headword “Date:” of “Date: Thursday, August 6, 2002” on the 8th line of the document in FIG. It is a rule to do. For example, rule B-1 is a rule for extracting a headword of “date-time-date”, and character strings such as “date-time:” and “date-date:” match.
[0037]
The extraction rule storage unit 3 also stores date and time classification rules using headwords, as shown in FIG. Rules C-1 to C-3 are date and time classification rules using headwords. For example, in the 8th line of the document in FIG. 3A, the <DATE> tag is set for the date “August 6, 2002 (Thursday)”, and the </ HD DATE OPEN> tag is set for the headword “Date:”. If is set, it becomes as follows.
<HD DATE OPEN> Date: </ HD> <DATE> August 6, 2002 (Thursday) </ DATE>
Here, since “August 6, 2002 (Thursday)” is behind the headword “Date:”, it is considered to be the date of the event. Therefore, rule C-1 is applied to classify this date into the date of the event.
[0038]
The extraction rules stored in the extraction rule storage unit 3 described above are not limited to the example shown in FIG. For example, the regular expression described in FIG. 10 of Japanese Patent Laid-Open No. 09-269940 and the format described in FIG.
[0039]
The result of extracting the document of FIG. 3A using the rules of FIG. 2 is as shown in FIG. In FIG. 3B, “date-time-date” is enclosed in a <DATE OPEN> tag, and “date-time-deadline” is enclosed in a <DATE LIMIT> tag. Also, “Event-Conference” is an <EVT CONFERENCE> tag, “Action-Participation Request” is an <ACT JOIN> tag, “Action-Response Request” is an <ACT ASK> tag, and “Location” is <POS> Each is surrounded by tags.
[0040]
(Document Information Registration Unit 4)
When the date and time of the character string extracted by the extraction unit 2 is a relative time such as “today” or “tomorrow”, the document information registration unit 4 sets the date and time to an absolute time. change. Also, morphological analysis is performed. For example, the first verb of the sentence including “First Conference Room” on line 9 in FIG. Judge that it is a venue. Then, the analysis results of these documents are registered in the extracted information storage unit 5.
[0041]
(Extracted information storage unit 5)
The extracted information storage unit 5 stores words appearing in the document as a result of document analysis, together with classifications such as date and time. FIG. 4 shows an example of data in the extracted information storage unit 5. FIG. 4 shows an example of classification of the extraction results of FIG. The extracted information storage unit 5 includes document number items, serial number items, type items, external representation items, internal representation items, line position items, and sentence position items.
[0042]
The document number item is an identification number assigned to each document. The serial number item is a number for identifying each extracted word. The type item is a classification of extracted words, and is divided into a large classification and a small classification. The external expression item is an expression in which the extracted word appears in the document. The internal expression item has a value when the relative time such as “today” is set to an absolute time such as “2002/8/3”. The line position item contains the line number where the word appears in the document. The sentence position item contains information indicating what number sentence the word appears in. However, when the target document is mail, −1 is set as the line position and sentence position of the word in the subject.
[0043]
(Document storage unit 6)
The document storage unit 6 stores the document itself with a document number. The document number is associated with the document number item in the extraction information storage unit 5.
[0044]
(Condition input unit 7)
The condition input unit 7 receives operations such as input of search conditions from the user.
[0045]
(Display unit 8)
The display unit 8 displays an input screen and outputs search results.
[0046]
(Search unit 9)
The search unit 9 performs a search process on the search condition input from the condition input unit 7. The search unit 9 includes a candidate selection unit 10, a fitness calculation unit 11, and a candidate alignment unit 12.
[0047]
(Candidate selection unit 10)
The candidate selection unit 10 searches for documents that match the user search conditions as candidate documents. At this time, in consideration of errors in the extraction results of the extraction unit 2 and the document information registration unit 4, the search condition is corrected so that more documents are hit than the documents that hit the search condition set by the user.
[0048]
(Fitness calculation unit 11)
The fitness level calculation unit 11 calculates the level of fitness with the search condition using the type item, line position item, and sentence position item of the extracted information storage unit 5 from the document listed by the candidate selection unit 10. The major classification and minor classification of the search conditions give a high degree of conformity to records that match those between hit records and those that are close to the line position or sentence position between hit records.
[0049]
(Candidate alignment unit 12)
The candidate alignment unit 12 performs alignment in descending order of fitness.
[0050]
The information search device 20 according to the present embodiment is configured as described above. Next, the operation of the information search device 20 will be described. First, document information registration processing will be described, and then document information search processing will be described. FIG. 5 is a flowchart showing the registration processing operation of the information search apparatus 20. Here, as an example, the document registration process shown in FIG.
[0051]
<Registration process>
(Step S100)
The extraction unit 2 performs an extraction process on the document input from the document input unit 1. The extraction rule shown in FIG. 2 is used. The result shown in FIG. 3B is obtained for the mail in FIG.
(Step S110)
The document information registration unit 4 issues a document number as preprocessing.
(Step S120)
The document information registration unit 4 registers the document input from the document input unit 1 in the document storage unit 6.
(Step S130)
The document information registration unit 4 performs the process shown in the flowchart of FIG. 6 on the document obtained in step S100. Hereinafter, a description will be given with reference to the flowchart of FIG.
[0052]
(Step S1000)
Set the line number to 0. Set the sentence number to 0.
(Step S1010)
The processing from step S1020 to step S1100 is executed for all the lines of the document. When the process is completed for all the rows, the process is terminated. If there is an unprocessed line, the process goes to step S1020.
(Step S1020)
Unprocessed lines are processed and the line number is incremented by one. Increase the sentence number by one.
(Step S1030)
Divide the line to be processed into morphemes. For example, the ninth line in FIG. 3B is “<POS>”, “first conference room”, “</ POS>”, “de”, “do (verb)”, “mas”, and “.”. However, a range surrounded by tags (for example, “first meeting room”) is treated as one morpheme.
(Step S1040)
The processing from step S1050 to step S1100 is executed for all morphemes. If executed, the process returns to step S1010.
[0053]
(Step S1050)
If the morpheme to be processed is “. (Punctuation mark)”, go to step S1060. If it is a tag other than the <HD> tag (heading tag), the end tag is found, and the process goes to step S1070. Otherwise, the process returns to step S1040. For example, in the ninth line of FIG. 3B, when “<POS>” is found, the process proceeds to the end tag “</ POS>”, and the process goes to step S1070. If "de", "do", or "mas", it is neither a tag nor a punctuation mark, so nothing is processed and the process returns to step S1040. If ".", Go to step S1060.
[0054]
(Step S1060)
Increase the sentence number by one. The process returns to step S1040.
(Step S1070)
Check if a small classification is set for the tag. If not, go to step S1080 to determine the minor category. If set, go to step S1090.
[0055]
(Step S1080)
Determine the minor classification. Check what varieties or verbs exist in the morphemes after this tag. If the tag is a <DATE> tag, and the subsequent morpheme contains “held (sa variable noun)”, “do (verb)”, and “open (verb)”, the subclass is determined to be the date of the event. If there is a “Deadline”, it is judged as the deadline. If the tag is a <POS> tag, and the subsequent morpheme contains "hold (sa variable noun)", "do (verb)", and "open (verb)", the subclass is determined to be the venue. Go to step S1090. For example, the 9th line in FIG. 3B has “do (verb)” after “<POS>” to “</ POS>”, so “<POS>” to “</ POS>”. Is judged as the venue.
[0056]
(Step S1090)
If the tag is a <DATE> tag, calculate the absolute time. As a method, for example, the method disclosed in JP-A-10-69472 is used. Go to step S1100.
(Step S1100)
Data is registered in the extracted information storage unit 5. If registered, it will return to step S1040.
[0057]
<Search process>
Next, the search process will be described. FIG. 7 is a flowchart showing the search processing operation of the information search apparatus 20 according to this embodiment. The data in the extracted information storage unit 5 is assumed to be as shown in FIG. Hereinafter, description will be given with reference to this flowchart.
[0058]
In the present embodiment, in order to search under various conditions, for example, a search condition input screen as shown in FIG. 8 is displayed to the user. FIG. 8 includes a type column and a specific value column. The type column is a column for selecting the type of word to be searched. The type column includes major classifications such as “date and time”, “event”, “location”, and “action”, and minor classifications such as “date” and “deadline”. “Not specified” is a column for searching without specifying a small classification. The small category column has a check button so that the category can be designated as a search condition. In the following description, in order to indicate each check button, the notation “date-date-date” and “action-not specified” are used.
[0059]
When searching for “documents including the date of the event”, check “date and time-date of event”. To search for “documents that include August 3, 2002 as the date”, check “Date-Date” and enter “2002/8/3” in the value field to the right of it. To do. The concrete value field may be blank if it is not particularly necessary to include the concrete value in the search condition.
[0060]
In addition, if you want to search all dates and times without distinguishing the deadline date or the date of the event, check "Date and time-not specified". When searching for “documents including August 3, 2002 as date and time”, check “Date and time-all” and enter “2002/8/3” in the specific value column to the right of the check.
[0061]
An AND search is performed by checking multiple types. For example, when searching for “documents including seminars where the venue is XX dome and the date is August 3, 2002”, check three places as shown in FIG. Enter a specific value for “Location-Venue”.
[0062]
(Step S200)
The candidate selection unit 10 relaxes the conditions in consideration of extraction errors with respect to the search conditions input from the condition input unit 7 and increases the number of documents that hit the search.
Relax the conditions as follows:
(1) When the user has set a specific value, a search is performed using only the specific value in order to cope with omission of extraction of the large classification and the small classification.
(2) When the user has not set a specific value, in order to deal with the omission of extraction of the small category, the search is performed using only the large category so that the record without the small category is hit.
[0063]
For example, in FIG. 8, the condition “place-venue” = “XX dome” is specified, but from the data in the extracted information storage unit 5, all the external representations are “XX dome”. Make the record hit.
[0064]
In the search condition shown in FIG. 8, a document that satisfies the following (condition 1), (condition 2), and (condition 3) in the extracted information storage unit 5 is specified.
(Condition 1) There is a record whose type item is “date-time-date” and the internal representation is “2002/8/3”.
(Condition 2) There is a record whose type item is “event-seminar”.
(Condition 3) There is a record whose type item is “place-venue” and the external expression is “XX dome”.
[0065]
In this embodiment, the search condition is relaxed and the search is performed using the following search condition.
(Condition 1 ') There is a record whose internal representation is "2002/8/3".
(Condition 2 ') There is a record whose major classification item is "event".
(Condition 3 ') There is a record whose external expression is "XX dome".
[0066]
(Step S210)
The candidate selection unit 10 searches the extracted information storage unit 5. For example, in FIG. 9A, the records (serial numbers 1, 2, 6, 8, 9, 10, 11) indicated by the shaded pattern are hit.
[0067]
(Step S220)
The candidate selection unit 10 collects the hit records with the same document numbers and sets the document numbers satisfying all the conditions as the processing target. For example, in FIG. 9, the data of document number 1 does not satisfy (Condition 3 ') and is not processed. The data of document number 2 is a processing target because it satisfies all the conditions.
[0068]
(Step S230)
The fitness level calculation unit 11 calculates the fitness level with the search condition for the document to be processed in step S220. First, the degree of fitness between each record and the search condition is calculated. This is set so that the larger the major classification or minor classification of the record matches the search condition, the higher. FIG. 10 shows the set value of the fitness. Different calculations are performed when a specific value is set for the search condition and when it is not set. When a specific value is set in the search condition, the goodness of fit is set to 1 if the major classification and minor classification of the retrieved record match the retrieval condition. If only the major classification of the retrieved record matches the retrieval condition and the minor classification of the retrieved record is blank, the fitness is 0.9. If only the major classification of the retrieved record matches the retrieval condition, the fitness is set to 0.5. If the major classification of the retrieved record is different from the major classification of the search condition, the fitness is set to 0.1.
[0069]
On the other hand, if no specific value is set in the search condition, the fitness is set to 1 if the major and minor classifications of the retrieved records match the search condition. If only the major classification of the retrieved record matches the retrieval condition, and the minor classification of the retrieved record is blank, the fitness is 0.85. If only the major classification of the retrieved record matches the retrieval condition, the fitness is set to 0.5.
[0070]
For example, serial number 8 in FIG. 9 corresponds to the case where (condition 2 ') matches and no specific value is set. Since only the large classification matches the search condition and the small classification is different from the search condition, the fitness is 0.5. The serial number 10 corresponds to (condition 1 ') and corresponds to a case where a specific value is set. Since the major classification matches the search condition and the minor classification is blank, the fitness is 0.9. The serial number 11 corresponds to (condition 3 ') and corresponds to a case where a specific value is set. Since the major classification matches the search condition and the minor classification is blank, the fitness is 0.9. The calculation result of this fitness is shown in the “S230 fitness” column in FIG. 9B.
[0071]
(Step S240)
The fitness is calculated based on the relationship between the appearance positions of words. When there are multiple search conditions, the closer the sentence position of the matched record is, the higher the score is given. If the sentence position of the record is the same, the fitness is set to 1. If the sentence position is different among multiple records, the value α with the smallest sentence position difference (minimum distance) is calculated for each matched record.
Goodness of fit = 1-0.1α
The fitness is calculated as In FIG. 9, since the serial number 9 has the smallest difference between the serial number 10 and the sentence position of 1 (= 3-2), the fitness is 0.9 (= 1−0.1 × (3-2) ) The calculation result of this fitness is shown in the “S240 fitness” column in FIG. 9B.
[0072]
(Step S250)
For each record, the two goodnesses calculated in steps S230 and S240
β × (goodness determined in step S230) + (1−β) × (goodness determined in step S240)
And calculate the grand total. In FIG. 9, when β = 0.5,
Serial number 8 is 0.5 × 0.5 + 0.5 × 0.7 = 0.6
Serial number 9 is 0.5 × 0.5 + 0.5 × 0.9 = 0.7
Serial number 10 is 0.5 × 0.9 + 0.5 × 0.9 = 0.9
Serial number 11 is 0.5 × 0.9 + 0.5 × 0.8 = 0.85
It becomes. Since the serial number 8 and the serial number 9 match the same (condition 2 '), only the serial number 9 having a high degree of fitness is totaled. As a result, the fitness of document number 2 is 2.45 (= 0.7 + 0.9 + 0.85). In the case of the search condition of FIG. 8, since there are three conditions, considering that the maximum value of the fitness is 3, the document is relatively suitable for the condition.
[0073]
(Step S260)
The candidate sorting unit 12 sorts using the fitness for each document calculated in step S250.
[0074]
(Step S270)
The display unit 8 outputs the search result. An output example is shown in FIG.
[0075]
(Effects of the first embodiment)
As described above, according to the present embodiment, since the configuration does not have the extraction rule for each field in advance, it is possible to eliminate the process of selecting in advance what field the document is.
[0076]
Further, at the time of extraction, words representing the date, time, place, etc. in the document are not stored in a table format, and the respective position information is stored. At the time of retrieval, the degree of matching with the retrieval condition is calculated using the type of each word and the positional relationship between the words. By arranging and displaying to the user in descending order of the fitness, the user can determine whether the search result can be trusted.
[0077]
Also, in the process of extracting the type of word, it is distinguished whether the date and time is a deadline date or a holding date. In this way, for example, the user can make a search by distinguishing the event date and the deadline date. On the other hand, regardless of the type of word specified by the user as a search condition, the search condition is relaxed rather than the search condition set by the user to search for candidate documents. There can be no leakage.
[0078]
In the present embodiment, the processing for determining the type of word from the document and registering it in the extraction information storage unit 5 (registration processing) is performed by the document input unit 1, extraction unit 2, extraction rule storage unit 3, and document information. This is done by the registration unit 4. Further, a process for searching for a document (search process) is performed by the condition input unit 7, the display unit 8, and the search unit 9. When the extracted information storage unit 5 and the document storage unit 6 are created in advance, the configuration of the information search device 30 shown in FIG. 12 may be adopted. That is, the input unit 7, the display unit 8, and the search unit 9, which are components that perform the search process, and the extracted information storage unit 5 and the document storage unit 6 may be included.
[0079]
(Second Embodiment)
In the first embodiment, a general document search has been described. In the present embodiment, a case will be described in which electronic mail is received via a network such as the Internet, the received electronic mail is sequentially registered, and searched according to a user request.
[0080]
FIG. 13 is an explanatory diagram showing a system configuration of the information search system 200 according to the present embodiment. Compared to the configuration of the first embodiment, a communication unit 13, a communication unit 14, a server 100, a client 110, and a network 120 are added. Server 100 and client 110 are connected to network 120.
[0081]
(Server 100)
As shown in FIG. 13, the server 100 includes a document input unit 1, an extraction unit 2, an extraction rule storage unit 3, a document information registration unit 4, an extraction information storage unit 5, a document storage unit 6, The search unit 9 and the communication unit 13 are included. The search unit 9 includes a candidate selection unit 10, a fitness calculation unit 11, and a candidate alignment unit 12. The server 100 registers mail information for electronic mail received through the network 120. In addition, it receives a search request sent from the client 110 from the user, and sends a search result. The communication unit 13 receives and transmits such information through the network 120. In addition, since each component other than the communication unit 13 is substantially the same as that of the first embodiment, a duplicate description is omitted.
[0082]
(Client 110)
As illustrated in FIG. 13, the client 110 includes a condition input unit 7, a display unit 8, and a communication unit 14. The client 110 is a computer such as a personal computer or a mobile terminal such as a mobile phone. The client 110 accepts user search conditions, connects to the network 120 from the communication unit 14, inquires the server 100, and outputs search results. Note that the components other than the communication unit 14 are substantially the same as those in the first embodiment, and a duplicate description thereof is omitted.
[0083]
(Network 120)
The network 120 may be of any protocol, topology, type of transmission medium, etc. A typical example of the network 120 is the Internet.
[0084]
The document information registration process is executed when the communication unit 13 receives an electronic mail. As in the first embodiment, the processing after step S100 is executed for the received email, so that the email information is registered in the extracted information storage unit 5. The document information search process is executed when the user inputs, for example, the input screen shown in FIG. The input screen of FIG. 8 is displayed using CGI (Common Gateway Interface) or the like.
[0085]
FIG. 14 is a flowchart showing the operation of the information search system 40 according to this embodiment. This will be described below with reference to FIG.
(Step S300)
In the client 110, the condition input unit 7 transmits the input search condition to the communication unit 14.
[0086]
(Step S310)
The communication unit 14 transmits the input search condition to the server 100 via the network 120.
[0087]
(Step S320)
The server 100 receives the search condition by the communication unit 13 and executes the processes after step S200 in the first embodiment.
[0088]
(Step S330)
In the server 100, the communication unit 13 transmits the search result to the client 110 via the network 120.
[0089]
(Step S340)
The display unit 8 outputs the search result.
[0090]
(Effect of the second embodiment)
As described above, according to the present embodiment, the present invention is also applied to a case where an electronic mail is received via a network such as the Internet, the received electronic mail is sequentially registered, and searched according to a user request. it can.
[0091]
(Third embodiment)
In the first embodiment, only the words surrounded by the tags are registered in the extracted information storage unit 5. Therefore, there may be a case where a word that should originally be registered as an event or a place is not registered due to an incomplete extraction rule. In order to deal with such extraction omissions, in this embodiment, in addition to the words surrounded by tags, words such as nouns appearing in the document that cannot be specifically assigned a major classification or minor classification are also registered. To.
[0092]
As shown in FIG. 15, the information search device 50 according to the present embodiment is configured by replacing the extracted information storage unit 5 of the information search device 20 according to the first embodiment with a word information storage unit 15. Has been. In addition to the words surrounded by the tags, the word information storage unit 15 registers words such as nouns that appear in the document, such as no major classification or minor classification. Other constituent elements are substantially the same as the constituent elements of the information retrieval apparatus 20, and a duplicate description is omitted.
[0093]
Next, the operation of the information search device 50 according to this embodiment will be described.
In order to register the noun appearing in the document in the word information storage unit 15, the result of the morphological analysis in step S1030 is used. Changes to FIG. 6 are shown in FIG.
[0094]
First, the process of step S1050 of the first embodiment is changed as follows. (Step S1200)
If the morpheme to be processed is “. (Punctuation mark)”, go to step S1060. If it is a tag other than the <HD> tag (heading tag), the end tag is found, and the process goes to step S1070. If the morpheme is a noun, go to step S1210. Otherwise, the process returns to step S1040. For example, “Yamada (noun)” is registered in the third line of FIG. 3B, and “following (noun)” and “schedule (noun)” are registered in the fifth line.
[0095]
Step S1210 is performed as follows.
(Step S1210)
Data is registered in the word information storage unit 15 with the major classification as “word”. If registered, it will return to step S1040.
[0096]
In addition, the registration destination in step S1100 of the first embodiment is changed to the word information storage unit 15. As a result, the data in the word information storage unit 15 of the document in FIG. 3B is as shown in FIG. As shown in FIG. 17, “words” are added to the major classification, and “everyone”, “Yamada”, “below”, “schedule”, “everyone”, “participation”, “how”, It is registered in the word information storage unit 15 as a noun.
[0097]
(Effect of the third embodiment)
As described above, according to the present embodiment, a search omission is caused by being registered in the word information storage unit 15 as a noun even if it cannot be originally extracted as an event or place due to incomplete extraction rules. Can be eliminated.
[0098]
The preferred embodiments of the information retrieval apparatus and information retrieval system according to the present invention have been described above with reference to the accompanying drawings, but the present invention is not limited to such examples. It will be obvious to those skilled in the art that various changes or modifications can be conceived within the scope of the technical idea described in the claims, and these are naturally within the technical scope of the present invention. It is understood that it belongs.
[0099]
For example, the following application examples can be considered.
(1) The degree of conformity shown in FIG. 10 can be set according to the large classification. Further, the calculation method of steps S240 and S250 is not limited to this.
(2) In step S210 of FIG. 7, a search may be made by partial matching of specific values. In addition, in the table of FIG. 10, the degree of suitability when the specific values partially match may be added to “when there is a specific value in the search condition”.
(3) Major classification and minor classification are not limited to those mentioned in the above embodiment. For example, a large classification such as a person name or an organization name is possible.
(4) The candidate sorting unit 12 does not necessarily need to sort in the order of suitability. Sorted in order from the date closest to the current date, the degree of fitness may be displayed when the list is displayed.
(5) Voice input may be performed instead of the input screen as shown in FIG.
(6) Since the sentence number is incremented by 1 for each line in step S1020 of FIG. 6, the sentence number cannot be calculated correctly when there is a line break in the middle of one sentence. Therefore, as pre-processing in step S100, processing may be performed so that line breaks do not occur in the middle of one sentence.
(7) With respect to the date and time conditions shown in FIG. 8, the candidate documents may be searched by expanding to a range of one week before and after the specified date and time. Also, the input method of FIG. 8 may be configured to allow the period to be input.
(8) The storage target of the document storage unit 6 may not be the document itself, but may be, for example, information (for example, URL) indicating the storage location of the document.
[0100]
【The invention's effect】
As described above, according to the present invention, it is possible to eliminate the process of selecting in advance what field the document is in by having a configuration that does not have an extraction rule for each field in advance.
[0101]
Further, at the time of extraction, words representing the date, time, place, etc. in the document are not stored in a table format, and the respective position information is stored. At the time of retrieval, the degree of matching with the retrieval condition is calculated using the type of each word and the positional relationship between the words. By arranging and displaying to the user in descending order of the fitness, the user can determine whether the search result can be trusted.
[0102]
Also, in the process of extracting the type of word, it is distinguished whether the date and time is a deadline date or a holding date. In this way, for example, the user can make a search by distinguishing the event date and the deadline date. On the other hand, regardless of the type of word specified by the user as a search condition, the search condition is relaxed rather than the search condition set by the user to search for candidate documents. There can be no leakage.
[0103]
The present invention can also be applied to a case where electronic mail is received via a network such as the Internet, the received electronic mail is sequentially registered, and searched according to a user request.
[0104]
In addition, even if it cannot be extracted as an event or place due to incomplete extraction rules, it is possible to eliminate omissions by registering it as a noun in the word information storage unit.
[Brief description of the drawings]
FIG. 1 is an explanatory diagram illustrating a configuration of an information search device according to a first embodiment;
FIG. 2 is an explanatory diagram showing data in an extraction rule storage unit;
FIG. 3 is an explanatory diagram illustrating an example of a document extraction result;
FIG. 4 is an explanatory diagram showing data in an extraction information storage unit;
FIG. 5 is a flowchart showing an operation of a registration process in the first embodiment.
6 is a flowchart showing details of step S130 in FIG. 5. FIG.
FIG. 7 is a flowchart showing an operation of search processing in the first embodiment.
FIG. 8 is an explanatory diagram showing an example of a search condition input screen.
FIG. 9 is an explanatory diagram showing data in an extraction operation storage unit;
FIG. 10 is an explanatory diagram showing a set value of fitness.
FIG. 11 is an explanatory diagram illustrating an example of search result output;
FIG. 12 is an explanatory diagram illustrating another configuration of the information search device according to the first embodiment;
FIG. 13 is an explanatory diagram showing a configuration of an information search system according to a second embodiment.
FIG. 14 is a flowchart showing the operation of the information search system in the second embodiment.
FIG. 15 is an explanatory diagram illustrating a configuration of an information search device according to a third embodiment;
FIG. 16 is a flowchart showing details of step S130 in FIG. 5;
FIG. 17 is an explanatory diagram showing data in a word information storage unit;
[Explanation of symbols]
1 Document input part
2 Extraction unit
3 Extraction rule storage
4 Document Information Registration Department
5 Extracted information storage unit
6 Document storage
7 Condition input part
8 Display section
9 Search part
10 Candidate selection section
11 Conformity calculator
12 Candidate alignment part
13 Communication Department
14 Communication Department
15 Word information storage
20 Information retrieval device
30 Information retrieval device
40 Information retrieval system
50 Information retrieval device
100 servers
110 clients
120 network

Claims (15)

In an information retrieval device for retrieving a document,
A document storage unit for storing documents;
As information for searching the document, an extraction information storage unit that stores character strings in the document for each classification;
A condition input section in which a character string and the classification can be input;
A candidate selection unit that selects a candidate document by searching a document from the document storage unit using a character string input to the condition input unit as a search condition , and the selected candidate document and the condition input unit A search unit including a goodness-of-fit calculation unit for calculating goodness of fit using the classified classification ;
With
The fitness level calculation unit assigns a first fitness level to candidate documents with the same classification among the candidate documents, and a fitness level lower than the first fitness level for a candidate document with the classifications not matching. An information retrieval device characterized by providing The candidate selection unit selects the candidate document using the classification as a search condition when only the classification is input without inputting a character string into the condition input unit,
The relevance level calculation unit applies a lower relevance to the candidate document selected by the candidate selection unit than the first relevance level when no character string is input to the condition input unit and only the classification is input. The information search device according to claim 1, wherein a degree is given . The classification is divided into at least a major classification and a minor classification,
The fitness calculation unit uses the selected candidate document and the major classification and minor classification input to the condition input unit, and the major classification matches in the candidate document , and the minor classification classification the second fit was applied to the candidate documents that match, the only major category is characterized by imparting a lower fitness than the second adaptation degree to the candidate documents that match, to claim 1 The information retrieval device described. The classification is divided into at least a major classification and a minor classification,
The candidate selection unit selects the candidate document by using the major classification as a search condition when a character string is not input to the condition input unit and the major classification and minor classification are input .
The relevance calculation unit uses the selected candidate document and the small classification input to the condition input unit, and among the candidate documents, a third relevance degree is applied to a candidate document that matches the small classification. the grant, characterized in that said imparting less fit than the third adaptation degree to the candidate document the small classification do not coincide, the information retrieval apparatus according to claim 1. An extraction unit for extracting a predetermined character string included in the document;
When the date information is included in the document, the extraction unit classifies the date information as either an event date or a deadline date based on an appearance position relationship between the date information and a predetermined character string. The information retrieval device according to claim 1, wherein the information retrieval device is stored in the extracted information storage unit.   The character string classification in the extraction information storage unit is composed of a large classification determined according to the extraction rules stored in the extraction rule storage unit and a small classification obtained by subdividing the large classification The information search device according to any one of claims 1 to 5.   The major classification further has a classification for registering a noun in the document stored in the extraction information storage unit regardless of the extraction rule stored in the extraction rule storage unit. An information retrieval device according to any one of claims 3, 4 and 6. In an information retrieval system that retrieves documents,
A server unit and a terminal unit connected via a network;
The server part
A communication unit that transmits and receives a character string and classification input to the condition input unit of the terminal unit and a search result with the terminal unit ;
A document storage unit for storing documents;
As information for searching the document, an extraction information storage unit that stores character strings in the document for each classification;
A candidate selection unit that searches a document from the document storage unit and selects a candidate document by using a character string input to the condition input unit of the terminal unit as a search condition, and uses the selected candidate document and the classification A search unit including a fitness calculation unit that calculates the fitness by
With
The terminal part is
A condition input section in which a character string and the classification can be input;
A communication unit that transmits and receives a character string and classification input to the condition input unit, and a search result;
With
The fitness level calculation unit assigns a first fitness level to candidate documents with the same classification among the candidate documents, and a fitness level lower than the first fitness level for a candidate document with the classifications not matching. An information retrieval system characterized by providing   9. The information retrieval system according to claim 8, wherein the document is an electronic mail exchanged via the network. The candidate selection unit selects the candidate document using the classification as a search condition when only the classification is input without inputting a character string into the condition input unit,
The relevance level calculation unit applies a lower relevance to the candidate document selected by the candidate selection unit than the first relevance level when no character string is input to the condition input unit and only the classification is input. The information search system according to claim 8 or 9, wherein a degree is given . The classification is divided into at least a major classification and a minor classification,
The fitness calculation unit uses the selected candidate document and the major classification and minor classification input to the condition input unit, and the major classification matches in the candidate document , and the minor classification classification the second fit was applied to the candidate documents that match, the only major category is characterized by imparting a lower fitness than the second adaptation degree to the candidate documents that match, claim 8 or 9. The information search system according to 9 . The classification is divided into at least a major classification and a minor classification,
The candidate selection unit selects the candidate document by using the major classification as a search condition when a character string is not input to the condition input unit and the major classification and minor classification are input .
The relevance calculation unit uses the selected candidate document and the small classification input to the condition input unit, and among the candidate documents, a third relevance degree is applied to a candidate document that matches the small classification. The information retrieval system according to claim 8 , wherein a fitness level lower than the third fitness level is given to a candidate document that does not match the small classification . An extraction unit for extracting a predetermined character string included in the document;
When the date information is included in the document, the extraction unit classifies the date information as either an event date or a deadline date based on an appearance position relationship between the date information and a predetermined character string. The information search system according to claim 8, wherein the information search system is stored in the extracted information storage unit.   The character string classification in the extraction information storage unit is composed of a large classification determined according to the extraction rules stored in the extraction rule storage unit and a small classification obtained by subdividing the large classification An information retrieval system according to any one of claims 8 to 13.   The major classification further has a classification for registering a noun in the document stored in the extraction information storage unit regardless of the extraction rule stored in the extraction rule storage unit. 15. An information retrieval system according to any one of claims 11, 12, and 14.

Images