JP2019008573A - Similar document retrieval device, similar document retrieval method and program - Google Patents

Abstract

To provide a document retrieval technique that does not need manual works with respect to documents such as preparations of impression notes, allocations of tags, and the like.SOLUTION: Assuming that documents are composed of a plurality of segments to be obtained through dividing of the document by a prescribed unit, and information attribute frequency series information on a document is information representing a series of frequency of an attribute corresponding to a word included in each of the plurality of segments composing the document, a similarity document retrieval device comprises: an input 71 that receives an input of attribute frequency series information; and a display unit 7 that displays information about the document with the input attribute frequency series information, and the attribute frequency series information high in similarity as a similar document retrieval result.SELECTED DRAWING: Figure 1

Description

  The present invention relates to a technique for searching for a document and a technique used for searching and displaying a document.

  There is a demand for searching for textbook-based documents such as “I want to read a happy-end book” or “Sometimes I will read a sad story”.

  Searching for a document such as a picture book based on the type of story can be realized by the following two methods.

  The first method is to search for a document such as a picture book based on a story type based on an impression sentence of the document created in advance. For example, it is possible to perform a search for matching impressions of documents using “happy ending” and “sad story” as keywords.

  The second method is to manually add tags such as “happy ending” and “sad story” to a document in advance and perform a search using this tag as a keyword (see, for example, Non-Patent Document 1). ).

Naruto University of Education Library, “Picture Book Database”, [Search April 14, 2017], Internet <URL: http://www.naruto-u.ac.jp/library/jidou/005.html>

  However, the above-mentioned first method has a problem that the search cannot be performed if the picture book does not have an impression sentence because it is not released or popular. In addition, the second method has a problem that the burden of manually attaching tags is large.

  An object of the present invention is to provide a similar document search apparatus, method, and program that do not require manual work on a document such as creation of a comment sentence or tag assignment.

  In the similar document search device according to one aspect of the present invention, a document is composed of a plurality of segments obtained by dividing the document into predetermined units, and words included in each of the plurality of segments constituting the document The attribute frequency series information as an attribute frequency series information, and an input unit that accepts input of attribute frequency series information selected or drawn by the user, and an attribute having a high degree of similarity to the input attribute frequency series information And a display unit for displaying information on the document having frequency series information as a similar document search result.

  There is no need for manual work on the document, such as creating an impression sentence and assigning a tag.

The block diagram which shows the example of a similar document search apparatus. The flowchart which shows the example of the similar document search method. The figure which shows the example of attribute frequency series information. The figure which shows the example of attribute frequency series information. The figure for demonstrating the example of normalization of a histogram. The figure for demonstrating the example of normalization of a histogram. The figure for demonstrating the example of normalization of a histogram. The figure for demonstrating the example of selection of attribute frequency series information. The figure for demonstrating the example of selection of attribute frequency series information. The figure for demonstrating the example of selection of attribute frequency series information. The figure for demonstrating the example of description of attribute frequency series information. The figure which shows the example of a similar document search result. The figure which shows the example of a similar document search result. The figure for demonstrating the example of the setting of a predetermined unit or the number of predetermined bins. The figure for demonstrating the example of the setting of a predetermined unit or the number of predetermined bins. The figure for demonstrating the example of the setting of a predetermined unit or the number of predetermined bins. The figure for demonstrating the example of the setting of a predetermined unit or the number of predetermined bins. The figure which shows the example of a similar document search result. The figure for demonstrating the example of the setting of a predetermined unit or the number of predetermined bins.

[Embodiment]
Hereinafter, an embodiment of the present invention will be described with reference to the drawings.

  FIG. 1 shows a configuration example of a similar document search apparatus. As shown in FIG. 1, the similar document search device includes a word extraction unit 1, a word attribute storage unit 2, an attribute frequency series information generation unit 3, an attribute frequency series information storage unit 4, a similarity calculation unit 5, and a similar document information output. For example, a unit 6 and a display unit 7 are provided. The display unit 7 includes a granularity setting unit 71 and an input unit 72.

  The similar document search method is realized, for example, when each unit of the similar document search apparatus performs the processing from step S1 to step S7 shown in FIG.

  As shown in FIG. 1, the similarity calculation apparatus 20 includes, for example, a word extraction unit 1, a word attribute storage unit 2, an attribute frequency series information generation unit 3, an attribute frequency series information storage unit 4, and a similarity calculation unit 5. Yes.

  The similarity calculation method is realized by, for example, each unit of the similarity calculation apparatus performing the processing from step S1 to step S5 illustrated in FIG.

<Word extraction unit 1>
The word extraction unit 1 takes a document as an input, extracts a word included in each of a plurality of segments constituting each input document (step S1), and outputs the extracted word to the attribute frequency series information generation unit 3 .

  The document may be a paper document such as a book, magazine, or newspaper in which characters or pictures are written, or may be an electronic document such as a web page, electronic book, or source code. An example of a document is a picture book. An example of an electronic document is a structured electronic document described in a markup language such as XML.

  It is assumed that the document is composed of a plurality of segments obtained by dividing the document into predetermined units. The predetermined unit is a unit constituting a document such as a sentence, a page, a paragraph, and a chapter. In this way, the predetermined unit may be a break due to display restrictions such as pages, etc., or a paragraph or chapter where the author intentionally breaks the story (the chapter summarizes the story development by the author) May be).

  When the document is a structured electronic document described in a markup language such as XML, the elements of the structured electronic document can be set as a predetermined unit. The predetermined unit may be a plurality of units constituting the document exemplified above. For example, the predetermined unit may be one page or two pages. As will be described later, the granularity setting unit 71 may set a predetermined unit size.

  For example, when the document is a picture book, a page that is an example of a break due to display restrictions may be used as a predetermined unit. This is because picture books are content in which pictures and characters are used in a complementary manner, and are often designed with a page as one unit, so that story development can be easily expressed by using a page as a predetermined unit. It is to become.

  When a picture book that is written for a relatively young age and is designed with two spread pages as one unit is a document, two pages may be set as a predetermined unit.

  In addition, when the document is a relatively long and complicated document such as a children's book or a novel, a chapter that is an example of a delimiter in which the author deliberately divides the story may be used as the predetermined unit. This is because, for such a document, it is better to use “delimiter where the author deliberately divides the story” as the predetermined unit than simply “delimitation due to display restrictions” as the predetermined unit. This is because it is easy to represent the development.

  For example, after extracting a word from a morphological analysis, a word is extracted from the result of the morphological analysis. Etc.). Independent words by dividing input documents by morphological analysis processing, extracting word strings by segmentation in parts of speech, adding parts of speech to each word, and extracting words corresponding to specific parts of speech And the cancellation expression that accompanies it.

When the predetermined unit is one page, words included in each page are extracted for each page of each document. For example, page 1 of a document contains the sentence “The war has begun. Carnivores have been killed.” Page 2 “Elephants do great tricks, but zoo staff cannot enjoy.” Page 3 , "The elephant died after all." In this case, for example, the following morphological analysis results can be obtained from page 1 by using the morphological analyzer from the sentence “The war has started. The carnivores have been killed.”
War nouns, weird transformations, *, *, *, *, wars, senso, senso are particles, case particles, general, *, *, *, ga, ga begins verbs, independence, *, *, five steps La line, continuous connection, beginning, hajima, hajimat, auxiliary verb, *, *, *, special, ta, basic form, ta, ta, ta. Symbol, punctuation, *, *, *, *. ,. ,.
Carnivorous nouns, savory connections, *, *, *, *, carnivorous, nikshoku, nikshoku beasts nouns, suffixes, general, *, *, *, beasts, ju, juju are particles, particle, *, *, * , Ha, wa killed verb, independence, *, *, five steps, sa line, form, kill, corosa, corrosa verb, suffix, *, *, single step, combined form, le, le auxiliary verb, *, *, *, Special / t, basic form, t, t, t. Symbol, punctuation, *, *, *, *. ,. ,.
EOS

Part-of-speech information (first two elements) and its standard form (sixth element) are obtained in comma-separated form after the appearance form of the word written from the original sentence. If the standard form of part-of-speech information corresponding to an independent word is extracted, the words “war”, “start”, “carnivorous”, “beast”, and “kill” are obtained. Also, from page 2, the following morphological analysis results can be obtained from the sentence “Elephant performs wonderfully, but zoo staff cannot enjoy it”.
Elephant noun, common noun, general, *, *, *, elephant, elephant, elephant, zo, elephant, zo, han, *, *, *, *
Is a particle, coordinator, *, *, *, *, ha, ha, wa, wa, wa, sum, *, *, *, *
Awesome adjective, general, *, *, adjective, community-general, subarusy, great, awesome, subashi, great, subaru, sum, *, *, *, *
Art noun, common noun, general, *, *, *, gay, art, art, game, art, game, Han, *, *, *, *
A particle, case particle, *, *, *, *, wo, a, o, a, o, sum, *, *, *, *
To verb, non-self-sustainable, *, *, sa line modification, ending form-general, sur, do, do, sur, do, sur, sum, *, *, *, *
Is particle, connective particle, *, *, *, *, ga, ga, ga, ga, ga, ga, sum, *, *, *, *
, Supplementary symbols, punctuation marks, *, *, *, * ,,,,,,,,, symbols, *, *, *, *
Animal noun, common noun, general, *, *, *, dove butterfly, animal, animal, dove butterfly, animal, dove butterfly, han, *, *, *, *
Soen Suffix, Noun, General, *, *, *, En, Soen, Soen, En, Soen, En, Han, *, *, *, *
Staff noun, common noun, general, *, *, *, schoin, staff, staff, schoin, staff, schoin, han, *, *, *, *
Is a particle, coordinator, *, *, *, *, ha, ha, wa, wa, wa, sum, *, *, *, *
Enjoy Verb, General, *, *, Lower 1st-Ma Line, Blank Form-General, Tanoshim, Enjoy, Enjoy, Tanoshime, Enjoy, Tanoshimeru, Sum, *, *, *, *
No auxiliary verb, *, *, *, auxiliary verb-nai, final form-general, nai, not, not, nai, not, nai, sum, *, *, *, *
. Auxiliary symbols, punctuation marks, *, *, *, * ,. ,. ,,. ,,symbol,*,*,*,*
EOS

In the same way as described above, the words "elephant", "great", "gei", "animal", "garden", "staff", and "enjoy" are extracted. Since there is an auxiliary verb "No", record "Enjoy (cancel)". Further, from page 3, for example, the following morpheme analysis result is obtained from the sentence “Elephant died after all”.
After all, noun, adverb possible, *, *, *, *, after all, sign, mark, symbol, punctuation, *, *, *, *, ...
Elephant noun, general, *, *, *, *, elephant, elephant, zo also particle, coordinator, *, *, *, *, m, mo dead verb, independence, *, *, five steps Line, continuous connection, die, thin, thin auxiliary verb, *, *, *, special, basic, basic, da, da. Symbol, punctuation, *, *, *, *. ,. ,.
EOS

  By the same method as described above, the words “finally”, “elephant” and “die” are extracted.

  It should be noted that the words “Enjoy” and “None” are obtained by morphological analysis as the words corresponding to the “Unable to enjoy” part of the sentence “The elephant does great art but cannot enjoy the zoo staff” on page 2. Suppose that The word “enjoy (cancel)” above corrects the appearance of the word “enjoy” and its role in the document because there is a cancellation verb “no” within 3 words from “enjoy”. A word that means that it is necessary. In this way, in the word string obtained by morphological analysis, negative expressions such as “no” and “bad” adjunctive auxiliary verbs within a predetermined number of words (three words in the above example) from a certain word A May appear as “word A (cancellation)”.

<Word attribute storage unit 2>
The word attribute storage unit 2 stores attributes corresponding to each of a plurality of words.

Examples of attributes are “positive” and “negative” polarities. In other words, the attribute corresponding to the word is information about whether the word gives a positive impression or a negative impression. For example, in a Japanese language evaluation polarity dictionary (for example, see References 1 and 2), an attribute “positive” or an attribute “negative” is assigned to each word. As the attribute of each word, the polarity of “positive” or “negative” assigned in the Japanese evaluation polarity dictionary can be used.
[Reference 1] Nozomi Kobayashi, Kentaro Inui, Yuji Matsumoto, Kenji Tateishi, Shunichi Fukushima, "Collection of Evaluation Expressions for Opinion Extraction", Natural Language Processing, Vol.12, No.3, pp.203-222, 2005.
[Reference 2] Masahiko Higashiyama, Kentaro Inui, Yuji Matsumoto, “Acquiring Noun Evaluation Polarity Focusing on Preferential Predicate Preference”, Proc. 14th Annual Conference of the Language Processing Society, pp.584-587, 2008.

  For example, the polarity “negative” is assigned as an attribute corresponding to the word “war”, the polarity “positive” is assigned as an attribute corresponding to the word “great”, and the word “enjoy” Assume that the polarity “positive” is assigned as the corresponding attribute, and the polarity “negative” is assigned as the attribute corresponding to the word “die”.

  As the attribute of the word, an attribute other than the two polarities such as “positive” and “negative” may be used. For example, by setting the attribute of a word that is not registered in the Japanese evaluation polarity dictionary or not stored in the word attribute storage unit 2 to “even”, three types of polarity “positive”, “negative”, and “even” May be used. Further, four types of polarities obtained by further subdividing the above two types of evaluation polarities such as “positive” and “negative” such as “joy”, “anger”, “sorrow”, and “easy” may be used. Further, the attribute of the word may be only one kind of polarity such as “positive” or “negative”. As described above, the word attributes may be one type or two or more types.

  Furthermore, information on whether or not a word is in a predetermined dictionary may be used as an attribute, such as information on whether or not it belongs to a word group described in the dictionary. For example, whether a certain word belongs to a predetermined word group that contributes to “development of the heart” regarded as important in developmental psychology or the like may be used as an attribute of the word. Words that contribute to mental development are words that relate to the emotions of children.

  In this case, a dictionary or list containing words that contribute to the development of the infant's heart is prepared, and a certain word belongs to a word group described in the dictionary or list, for example, “think” or “think”. Whether or not the information is an attribute.

See references 3 and 4 for words that contribute to mental development. For example, it is possible to use a list of words that contribute to the mental development of the infant described in References 3 and 4.
[Reference 3] Simon Baron-Cohen, Ofer Golan, Sally Wheelwright, Yael Granader, and Jacqueline Hill1, "Emotion Word Comprehension from 4 to 16 Years Old: A Developmental Survey", Frontiers in Evolutionary Neuroscience 2 (109): 109 November 2010
[Reference 4] Yayoi Watanabe, Saori Fujino, "Development of emotional literacy in children: focusing on emotional expression", Bulletin of Faculty of Letters, Hosei University (73), 83-97, 2016

  In this way, in the assignment of word attributes, there are industry dictionaries such as evaluation dictionaries such as the above-mentioned Japanese evaluation polarity dictionary and predetermined word groups that contribute to “development of the heart” that is regarded as important in the developmental psychology and the like. An existing word set such as a word group used by a person for reputation analysis may be used.

In addition, when the words are grouped so as to belong to any one of at least one different group, the group to which each word belongs can be used as an attribute. The grouping of each word can be performed by using a topic extraction method using a topic model described in Reference 5, for example.
[Reference 5] David M. Blei, Andrew Y. Ng, Michael I. Jordan, "Latent Dirichlet Allocation", JMLR2003

  In topic extraction using a topic model, by inputting a predetermined number of topics and a specific document set, words included in the input specific document set are divided into topics (= groups) having a predetermined number of topics. Can do. For example, by using as an attribute whether or not the subject belongs to a topic obtained by topic extraction using the topic model, an attribute of a word suitable for the input specific document set can be given. The topic extraction using the topic model and the assignment of the word attribute using the attribute obtained by the topic extraction are performed, for example, by the attribute assignment unit 8 indicated by a broken line in FIG. The similarity calculation device 20 and the similar document search device may further include an attribute assignment unit 8.

  In this way, assignment of word attributes suitable for a specific document set may be performed without using an existing word set.

  For example, a document set to be searched may be used as a specific document set. That is, some or all of the words included in the document to be searched are grouped by a predetermined grouping method so that each word belongs to at least one different group, and the group to which the word belongs or An attribute that does not belong to any group may be an attribute corresponding to the word. Thereby, it is possible to assign word attributes suitable for the document set to be searched.

  Hereinafter, <Method 1> <Method 2>, which is a method other than a method using a topic model for grouping so that each word belongs to any one of at least one different group using a specific document set, will be described. .

<Method 1>
Method 1 uses word2vec and clustering.

First, the attribute assignment unit 8 uses a specific document set as an input, and embeds words appearing in the document into the Euclidean space using word2vec (step G11). An example of a specific document set is a document set to be searched. For details on embedding words in the Euclidean space using word2vec, see, for example, Reference 6.
[Reference 6] Suzuki Jun, Nagata Masaaki, “Right-truncatable Neural Word Embeddings”, Proceedings of the 2016 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pp.1145--1151,

  Then, the attribute assignment unit 8 divides the word group embedded in the Euclidean space into k clusters (a word set or a word group) by clustering (step G12). k is a predetermined positive integer. For clustering, an existing method such as a k-means method or a random forest may be used. In the case of <Method 1>, the presence or absence of belonging to any one or a plurality of groups obtained by clustering may be used as an attribute.

<Method 2>
Method 2 uses so-called bag of words and clustering. Since a document is represented as a bag containing several important words, the vector construction described below is called a bag of words.

  First, the attribute assigning unit 8 examines all the C documents 1,..., C included in a specific document set for independent words appearing in the document, and creates an independent word dictionary appearing in the document (step G21). ). An example of a specific document set is a document set to be searched.

  Suppose that there are V types of free-standing words in the created dictionary. That is, it is assumed that the independent words appearing in the documents 1,..., C are 1,. The attribute assigning unit 8 counts the number of books of the document in which each of the V independent words appears (step G22). That is, the so-called df (document frequency) is examined for each word such that the independent word “wolf” appears in 20 documents, the independent word “xx” appears in the book document, and so on.

  The attribute assigning unit 8 determines a certain threshold value D and identifies independent words whose df is higher than the upper D rank (step G23). D is a predetermined positive integer. By truncating free words whose df is less than D, the calculation cost can be reduced.

  v = 1,..., V, and D is a positive integer determined in advance, and the number of documents in which the independent word v appears in the set of documents to be searched, obtained by the processing of step G22 and step G23. Let D, 1, ..., D be the independent words from the largest df (v).

  The attribute assigning unit 8 examines how many kinds of D words or more of independent words D obtained in step G23 appear in each document (step G24). That is, for example, the attribute assignment unit 8 checks that two types of independent words “Okami” at the df position and independent words “xx” at the df position appear in the document “Akazuki”.

  At this time, the attribute assigning unit 8 may reduce the amount of calculation thereafter by truncating lower-level documents with a small number of types of independent words appearing. That is, if the number of documents is small, it is not necessary to truncate.

  c = 1,..., C, and N is a positive integer determined in advance. Among the D independent words appearing in the document c included in the document set to be searched, obtained in the process of step G24 N documents from the one with the largest number of independent words M (c) are 1,. If N <C, this corresponds to truncating the lower N-C documents with a small number of independent word types. When N = C, this corresponds to not performing truncation.

  As a result of the processing in step G24, there remains a selection of independent words and documents. Remaining independent words and documents are set as independent word set V '= {1, ..., D} and document set E' = {1, ..., N}, and each independent word set V 'appears in many documents. In order, the document set E ′ is arranged in the order including many independent words.

  At this time, the attribute assigning unit 8 determines how many times each independent word appears in the documents in the document set E ′ as a feature vector of the independent word (step G25). For example, the self-supporting word “Wolf” is 15 times for “Akazukin” in the first place in the document, 10 times in “Okami and 7 Hiki-no-kogagi” in the second place in the document,…. If it appears once, the feature vector of the self-supporting word “wolf” becomes an | E ′ | -dimensional vector (15, 10,..., 0). Here, | E ′ | is the number of documents included in the document set E ′ (N in this case).

  d = 1,..., D, n = 1,..., N, and the number of times the independent word d appears in the document n is O (n, d), corresponding to the independent word d obtained by the process of step G25. The vector to be performed is (O (1, d),..., O (N, d)).

  By the processing from step G21 to step G25, each independent word is arranged in the | E '| dimension feature space. The attribute assigning unit 8 divides the independent words arranged in the | E ′ | -dimensional feature space into k clusters (a word set or a word group) (step G26). That is, the independent words 1,..., D are clustered by clustering the independent words 1,.

  k is a predetermined positive integer. The clustering method and how to treat each group as an attribute after clustering are the same as those described in step G12 of <Method 1>.

  A plurality of attributes may be assigned to one word.

<Attribute Frequency Series Information Generation Unit 3>
The attribute frequency series information generation unit 3 receives the words extracted by the word extraction unit 1.

  The attribute frequency series information generation unit 3 reads the attribute corresponding to the word extracted by the word extraction unit 1 from the word attribute storage unit 2, and sets the attribute corresponding to the word included in each of the plurality of segments constituting each document. By counting the frequency, attribute frequency series information which is information representing a series of attribute frequencies corresponding to words included in each of a plurality of segments constituting each document is obtained (step S3). The attribute frequency series information of each document output by the attribute frequency series information generation unit 3 is stored in the attribute frequency series information storage unit 4 in association with information for identifying each document and information for identifying a segment.

  When the predetermined unit is one page, the attribute frequency series information generation unit 3 counts the frequency of the attribute corresponding to the word included in each page.

  For example, assume that there are two types of attributes, “positive” and “negative”, and no attribute is assigned to a word that is not stored in the word attribute storage unit 2. Then, assuming that “War has started. Carnivores have been killed” is described on page 1, the word extracted by the word extraction unit 1 and input to the attribute frequency series information generation unit 3 is “ “War” “Beginning” “Carnivorous beast” “Kill”.

  When the word attribute storage unit 2 assigns the attribute “negative” to the word “war”, the attribute frequency series information generation unit 3 counts the frequency of the attribute “negative”. Specifically, the frequency of the attribute “negative” having an initial value “0” is added by “1”. If the word attributes “start”, “carnivorous”, and “kill” are not stored in the word attribute storage unit 2, they are not counted as the frequency of any of the “positive” and “negative” attributes. By this counting process, the frequencies of the attributes corresponding to the words included in page 1 are “positive”: 0 and “negative”: 1.

  The same processing is performed for page 2, page 3,. For example, the frequency of the attribute corresponding to the word included in page 2 is “positive”: 1, “negative”: 1, and the frequency of the attribute corresponding to the word included in page 3 is “positive”: 0, “negative”: For example, the frequency of attributes corresponding to words included in each page is counted for each page.

  Note that the attribute frequency series information generation unit 3 stores a word attribute storage for words with accompanying information indicating the necessity of correction, such as “enjoy (cancel)” extracted from the page 2 by the word extraction unit 1. You may count as the appearance of the word with the evaluation polarity which reversed the evaluation polarity found by referring to the part 2. For example, when the polarity “positive” is assigned as the attribute corresponding to the word “enjoy” in the word attribute storage unit 2, the attribute frequency series information generation unit 3 selects the attribute “enjoy (cancel)”. May count the frequency as “negative” with the polarity “positive” reversed. Alternatively, the attribute frequency series information generation unit 3 may regard the positive / negative evaluation polarity of the word as lost due to the cancellation expression, and may not count any evaluation polarity.

  Thus, the frequency of the attribute corresponding to the word included in each page counted for each page is an example of attribute frequency series information.

  Note that a document is composed of S segments 1,..., S, and the attributes corresponding to words are A attributes 1,..., A, and s = 1,. , A = 1,..., A, an index representing the frequency of the word t included in the segment s is Hs (t), and a set of words ts (a) whose corresponding attribute is included in the segment s is { ts (a)}, and the frequency Fs (a) of attribute a in segment s is Fs (a) = Σts (a) ∈ {ts (a)} Hs (ts (a)). The attribute frequency series information may be information representing ((F1 (1), ..., F1 (A)), ..., (FS (1), ..., FS (A)).

  Here, the index Hs (t) indicating the frequency of the word t included in the segment s may take a smaller value as the number of segments including the word t in the document increases. Specifically, the frequency of the word t included in the segment s is Hs (t) ′, and the number of segments including the word t in the document is f (t), and Hs (t) = Hs ( t) ′ × log (S / f (t)). Here, the base of the logarithm is a positive real number other than a predetermined one such as e or 10.

  For example, it is assumed that the words “Owami”, “Silver”, “Wolf”, “Owami”, “Sheep”, and “Owami” are extracted from the first segment 1 and input to the attribute frequency series information generation unit 3. . In addition, it is assumed that there are two types of attributes corresponding to words, “positive” and “negative”.

  In this case, the attribute frequency series information generation unit 3 counts the frequency of the word “Okami” included in page 1 and sets the frequency to “3” (step S31).

  Next, the attribute frequency series information generation unit 3 reads the attribute of the word “Okami” from the word attribute storage unit 2 (step S32). When the attribute “negative” is given to the word “Okami” in the word attribute storage unit 2, the attribute frequency series information generation unit 3 sets the attribute of the word “Okami” to “negative”.

  In this example, it is assumed that many pages other than page 1 include many words “Okami”. In this case, the number of “negative” increases due to the fact that many words “Okami” are included, and there is a possibility that accurate story development cannot be extracted. Therefore, in this example, the attribute frequency series information generation unit 3 considers weights that reduce the influence of words appearing over a plurality of segments. An example of the weight is log (S / f (Okami)), where f is the number of segments including the word “Okami”. This weight is an application of TF-IDF (Term Frequency-Inverse Document Frequency) used when characterizing a document, and can also be called TF-IPF (Term Frequency-Inverse Page Frequency). Of course, as the weight, a weight that takes a smaller value as the number of segments including the word t in a certain document other than TF-IPF may be used.

  The attribute frequency series information generation unit 3 multiplies “3”, which is the frequency of the word “Okami” included in page 1, by the above “log (S / f (Okami))” “3 × log (S / f (Okami)) ”is set as an index“ H1 (Okami) ”indicating the frequency of the word“ Okami ”included in page 1 (step S33). In addition, “Okami” in H1 (Okami) represents that the input of the function having the word H1 as an argument is the word “Okami”.

  As described above, the index Hs (t) indicating the frequency of the word t included in the segment s is, for example, a value obtained by correcting the frequency of the word t included in the segment s with a weight such as TF-IPF. Hs (t) may be the frequency of the word t included in the segment s.

  Since the attribute frequency series information generation unit 3 has “negative” as the attribute of the word “Okami”, in other words, the word “Okami” includes the word t1 () corresponding attribute included in page 1 is “negative”. Since it is included in the set of positive) {t1 (positive)}, H1 is added to the variable F1 (negative) representing the frequency of “negative” on page 1 (step S34).

  The attribute frequency series information generation unit 3 performs the processing from step S31 to step S34 on each word of page 1, thereby causing the frequency of attribute “negative” in page 1 to be F1 (negative) = Σt1 (negative) ∈ {t1 ( negative)} H1 (t1 (negative)) and frequency F1 (positive) = Σt1 (positive) ∈ {t1 (positive)} H1 (t1 (positive)) of the attribute “positive” in page 1 (step) S35). (F1 (positive), F1 (negative)) is an example of page 1 attribute frequency series information.

  The attribute frequency series information generation unit 3 performs the processing from step S31 to step S35 on each page, thereby obtaining attribute frequency series information ((F1 (positive), F1 (negative)),..., (FS (positive) ), FS (negative)) (step S36), for example, information representing ((F1 (positive), F1 (negative)), ..., (FS (positive), FS (negative)) obtained in this way. Is an example of attribute frequency series information.

  As described above, the attribute frequency series information may be a histogram representing a series of attribute frequencies corresponding to words included in each of a plurality of segments constituting the document. As described below, the attribute frequency series information may be an approximate curve of the histogram.

  Note that a series obtained by weighting and adding a series of indices indicating the frequency of attributes corresponding to words included in each of a plurality of segments constituting each document may be used as the attribute frequency series information. For example, an index indicating the frequency of an attribute a (i = 1,..., A) having a segment s (s = 1,..., S) is Hia, the weight of the attribute a is αa, and Hs = Σi = 1AαiHis. In this case, H1,..., HS are attribute frequency series information.

  For example, when the attribute is “positive” or “negative”, the weight of the attribute “positive” is “1”, and the weight of the attribute “negative” is “−1”.

  Further, the attribute frequency series information of the document is represented by a single curve by using the approximate curve of the attribute frequency series information H1, ..., HS generated by weighted addition of the attribute frequency series as the attribute frequency series information. May be. The attribute frequency series information generation unit 3 can obtain an approximate curve by, for example, performing spline interpolation on the attribute frequency series information H1,. By expressing the attribute frequency series information of a document with a single curve, the story development can be expressed in an easy-to-understand manner.

  3 and 4 show other examples of attribute frequency series information. FIG. 3 and FIG. 4 show approximate curves of frequency series corresponding to words included in “positive” and “negative” for each page. The attribute frequency series information generation unit 3 can obtain an approximate curve by, for example, performing spline interpolation on the frequency series corresponding to the words included in each of “positive” and “negative”. As described above, the attribute frequency series information may be an approximate curve of a histogram representing a series of attribute frequencies. 3 and 4, the horizontal axis represents the page number, the vertical axis represents the frequency, the solid line represents the appearance frequency of the word belonging to the “positive” attribute, and the broken line represents the appearance of the word belonging to the “negative” attribute. Represents the frequency.

  FIG. 3 is an example of attribute frequency series information of a document having a typical “Happy End” story development. In FIG. 3, there is a portion where “negative” exceeds “positive”, but since “positive” finally exceeds “negative”, it can be said that the story development of “Happy End” is represented.

  FIG. 4 is an example of attribute frequency series information of a document having a typical “sad story” story development. In FIG. 4, there is a part where “positive” exceeds “negative”, but “negative” exceeds “positive” in most parts, and “negative” exceeds “positive”. It can be said that the story development of "sad story" is expressed.

  Thus, the attribute frequency series information of a certain document represents the story development of that document. For this reason, it is possible to search for a document with similar story development by searching for a document with similar attribute frequency series information in the processing of the similarity calculation unit 5 and the similar document information output unit 6 described later.

  For example, the attribute frequency series information generation unit 3 may smooth the attribute frequency series information obtained by the above processing by applying a low-pass filter. For example, when generating the attribute frequency series information for a picture book that is written for a relatively young age and designed with two pages of spread as one unit, When the text is described on the left page of the two spread pages and only the picture is described on the right page of the two spread pages, the unevenness of the attribute frequency series information may be extreme. In such a case, the unevenness of the attribute frequency series information can be flattened by smoothing the attribute frequency series information. Thereby, the precision of the search of a similar document can be improved.

  The attribute frequency series information generation unit 3 determines whether or not attribute frequency series information has been generated for all documents to be processed. If there is a document for which attribute frequency series information has not yet been generated, The process of S1 and step S3 is performed. That is, the processing of step S1 and step S3 is repeatedly performed until the attribute frequency series information is generated for all the documents to be processed.

<Attribute Frequency Series Information Storage Unit 4>
The attribute frequency series information storage unit 4 stores the attribute frequency series information generated by the attribute frequency series information generation unit 3 in association with the document identification number (ID or the like).

  As will be described later, the attribute frequency series information generated by the attribute frequency series information generation unit 3 may be stored in association with a document identification number (ID or the like) and a predetermined unit.

  The processing of the word extraction unit 1 and the attribute frequency series information generation unit 3 is, in other words, the processing of storing the attribute frequency series information of each document in the attribute frequency series information storage unit 4 is similar to the similarity calculation unit 5 and the similar document information. This is performed prior to the processing of the output unit 6 and the display unit 7. Unless the document set to be searched is changed, it is not necessary to redo the process of storing the attribute frequency series information of each document in the attribute frequency series information storage unit 4.

<Similarity calculation unit 5>
The similarity calculation unit 5 receives attribute frequency series information of a document that is a search query selected or depicted by the user through the display unit 7. The input of the attribute frequency series information of the document to be a query by the user through the display unit 7 will be described later.

  When the attribute frequency series information of the document serving as the search query is depicted by the user, the attribute frequency series information of the document serving as the search query depicted by the user is directly input to the similarity calculation unit 5. The

  When the user selects a document to be a search query, the similarity calculation unit 5 determines whether the attribute frequency series information storage unit 4 stores the attribute frequency series information of the document selected by the user. Determine whether. When the attribute frequency series information of the document selected by the user is stored in the attribute frequency series information storage unit 4, the similarity calculation unit 5 displays the attribute frequency series information of the document selected by the user as the attribute frequency series. Read from the information storage unit 4. If the attribute frequency series information of the document selected by the user is not stored in the attribute frequency series information storage unit 4, the steps by the word extraction unit 1 and the attribute frequency series information generation unit 3 for the document selected by the user The processing of S1 and step S3 is performed, and the attribute frequency series information of the document selected by the user obtained by this processing is input to the similarity calculation unit 5. In this case, the attribute frequency series information storage unit 4 stores the attribute frequency series information of the document selected by the user, which is obtained by this process, so that the attribute frequency series information storage unit 4 stores the attributes. The database of frequency series information may be updated.

  The similarity calculation unit 5 calculates the similarity between the attribute frequency series information of each document read from the attribute frequency series information storage unit 4 and the input attribute frequency series information (step S5). The calculated similarity is output to the similar document information output unit 6 as a set together with the document identification information.

  The comparison of the relationship between two attribute frequency series information has been explained so far with similarities, but the similarities include similarity such as Bhattacharyya Coefficient, Kullback-Leibler divergence, Itakura-Saito divergence, Bregman divergence, etc. A distance scale such as Dynamic Time Warping and Earth mover's distance when Euclidean distance is used as the dissimilarity of, and local distance can be used. Of course, as long as the similarity between two pieces of attribute frequency series information can be measured, an index representing similarity other than the above may be used.

  When the attribute frequency series information is a histogram representing an attribute frequency series, the number of bins in the histogram that is attribute frequency series information of two documents for calculating the similarity between them may be different. For example, when the predetermined unit size is one page and the number of pages of these two documents is different, the number of histogram bins that are attribute frequency series information of these two documents is different. become.

  Here, when using an index based on the assumption that the number of bins is the same, such as Bhattacharyya Coefficient, the number of histogram bins that are attribute frequency series information of two documents must be the same. There is.

  In this case, the similarity calculation unit 5 calculates the similarity between each other so that the number of bins in the histogram, which is the attribute frequency series information of the two documents for calculating the similarity between each other, becomes the predetermined number of bins. The histogram which is the attribute frequency series information of the two documents to be normalized is normalized, and the similarity is calculated based on the normalized histogram. The predetermined number of bins may be a predetermined fixed value (for example, “4” assuming “start”, “approval”, “turn”, “conclusion”), and calculate the similarity between each other. The number of bins of the histogram that is the attribute frequency series information of the two documents to be used may be the smaller number of bins, or the histogram of the attribute frequency series information of the two documents that calculate the similarity between the two documents It may be the number of bins with the larger number of bins, or may be the greatest common divisor of the number of bins in the histogram that is attribute frequency series information of two documents for calculating the similarity between each other. The predetermined number of bins may be the number of bins set by a granularity setting unit 71 described later.

  Hereinafter, an example of histogram normalization will be described. The number of bins of the histogram that is the attribute frequency series information of the document A shown in FIG. 5 is 4, and the number of histograms that is the attribute frequency series information of the document B is 6. The horizontal axis of the graph of FIG. 5 is the bin, and the vertical axis of the graph of FIG. 5 is the ratio of the total frequency.

  Consider normalizing these histograms so that the number of bins is four. Since the number of bins in the histogram of document A is originally four, nothing is done about the histogram of document A.

  On the other hand, since the number of bins in the histogram of document B is 6, for example, as shown in FIG. 6, it is necessary to distribute the frequency for each bin having a size of 6/4 = 1.5. In FIG. 6, the second bin and the fifth bin are located at the sorting boundary by sorting by the 1.5-size bin. As described above, the frequency of the bins located at the distribution boundary is determined by, for example, proportional distribution. In this example, the frequency of the second bin and the fifth bin is distributed by 50%. As a result, the histogram of document B is as shown in FIG. For example, histogram normalization may be performed in this way.

  Hereinafter, an example of selection of attribute frequency series information of a document to be a search query by the user will be described.

  It is assumed that the display shown in FIG. 8 is displayed on the display unit 7 which is a display device such as a liquid crystal display or a CRT. The user inputs information of a document to be a search query into an input device such as a keyboard or a touch panel in a text box for inputting a title or an author at the upper left in FIG. In the example of FIG. 8, the title “Irish Giraffe” is entered in the title text box.

  In this state, when the user clicks a “search” button with a pointing device such as a mouse or a touch panel, a search device (not shown) performs a search process associated with the “search” button at the clicked position. That is, a search process such as a forward match search is performed based on the information input in the text box, and the search result is displayed in the right column of FIG. In the example of FIG. 8, as a search result of a document starting with the character string “poor giraffe”, one item “poor giraffe” is displayed as the search result. When the user selects the search result “poor giraffe” with the pointing device, the display unit 7 displays FIG.

  In FIG. 9, on the search result display screen, two buttons of “picture book with similar characters and events” and “picture book with similar story development” are further displayed in a pop-up manner. Here, when the user clicks “picture book with similar story development” using a pointing device, the title “poor giraffe” is input to the similarity calculation unit 5 of the similar document search apparatus as a query document.

  The similarity calculation unit 5 reads the attribute frequency series information of the document having the title “poor giraffe” from the attribute frequency series information storage unit 4 with the title “poor giraffe” as an input, and searches for the document that is the search query. It is attribute frequency series information. If the attribute frequency series information of the document titled “Unhappy Giraffe” is not stored in the attribute frequency series information storage unit 4, the word extraction unit 1 and the attribute frequency series are selected for the document titled “Kairin Kirin”. The processing of the information generation unit 3 is performed, and the attribute frequency series information obtained by this processing is input to the similarity calculation unit 5.

  When a user clicks on a picture book with similar characters and events using a pointing device, a picture book with similar characters and events to a document titled “Kairin Kirin” is the similar document search device according to the present invention. A search is performed by a different document search apparatus, and the search result is displayed. An example of the document search device is an information presentation device described in, for example, Japanese Patent Application Laid-Open No. 2014-235723 or Japanese Patent Application Laid-Open No. 2006-148927.

  The information presentation apparatus disclosed in Japanese Patent Application Laid-Open No. 2006-148927 is used to display characters such as “Kirin” and “Zoo” from the input query (in the above example, a document titled “Kairin Kirin”) and the stage of a picture book. Bibliographic information such as feature words and authors in the document to be represented is extracted as features, and includes features whose similarity to the query features exceeds a predetermined value (that is, a common character or stage appears, or is written by the same author) A plurality of pieces of object information (for example, picture books) having a common point such as a picture book. In the information presentation apparatus disclosed in Japanese Patent Application Laid-Open No. 2006-148927, the higher the similarity between the features of the object information indicated by each node, the higher the similarity between the two-dimensional plane coordinates of each node indicating each of the searched object information Calculate so that the distance between them is closer, calculate the altitude of the node that is higher as the similarity between the object information and the query is higher, and the node at the 3D coordinate position indicated by the calculated 2D plane coordinates and height information Are plotted, and an object plotted in a predetermined range from the query is presented as a search result.

  The display screen of the display unit 7 illustrated in FIGS. 8 and 9 is an example of the input unit 72 that receives input of attribute frequency series information selected by the user.

  Hereinafter, another example of selection of attribute frequency series information of a document to be a search query by the user will be described.

  Assume that the display unit 7 displays FIG. In FIG. 10, an example of attribute frequency series information representing a happy ending is shown on the left side of FIG. 10, and an example of attribute frequency series information representing a sad story is shown on the right side of FIG. In FIG. 10, the attribute frequency series information is represented by a single curve, the horizontal axis represents the degree of progress of the story, and the vertical axis indicates that the positive degree increases as it goes up and the negative degree decreases as it goes down. Suppose it means a good score. When the user selects the happy end graph on the left side of FIG. 10 with the pointing device, the attribute frequency series information corresponding to the happy end graph is selected. When the user selects the sad story graph on the right side of FIG. 10 with the pointing device, the attribute frequency series information corresponding to the sad story graph is selected.

  The display screen of the display unit 7 illustrated in FIG. 10 is an example of the input unit 72 that receives input of attribute frequency series information selected by the user.

  Hereinafter, an example of description of attribute frequency series information of a document to be a search query by the user will be described. In FIG. 11, the attribute frequency series information is represented by a single curve. By expressing the attribute frequency series information by a single curve, the attribute frequency series information can be input more intuitively by the user than by expressing the attribute frequency series information by two or more curves. It is possible to reduce the burden on the user's input of depicting the image.

  When the user clicks the drawing area of the curve in FIG. 11 with a pointing device, a vertex can be designated. In FIG. 11, the vertices are indicated by ◯. When the user designates a plurality of vertices, a curve such as an nth-order spline curve passing through the designated vertices is generated. A new vertex can be specified by clicking on the curve, and the shape of the curve may be changed by dragging the vertex. For example, a curve representing attribute frequency series information is drawn in this way.

  The display screen of the display unit 7 illustrated in FIG. 11 is an example of the input unit 72 that receives input of attribute frequency series information drawn by the user.

  Also, in FIG. 11, an example of a curve corresponding to attribute frequency series information of a document having a “happy ending” story, an attribute frequency series information of a document having a “sad story” story, beside the curved drawing area. An example of the corresponding curve is displayed. In this way, by displaying an example of a curve corresponding to the attribute frequency series information of a document having each story, it can be used as a model when the user draws a curve, so that the user can easily draw the curve. Become.

  For example, the attribute frequency series information that is a curve drawn in this way may be input as it is to the similarity calculation unit 5 as the attribute frequency series information of a document that is a search query, or the attribute frequency series information that is a curve May be converted into a histogram having a predetermined number of bins, and the converted histogram may be input to the similarity calculation unit 5 as attribute frequency series information of a document to be a search query.

  Here, the attribute frequency series information of a certain document represents the story development of the document. For this reason, documents with similar attribute frequency series information are highly likely to be documents with similar stories. For this reason, a document with a similar story can be found by finding a document with similar attribute frequency series information. Therefore, the similarity calculation unit 5 calculates the similarity of the attribute frequency series information.

<Similar Document Information Output Unit 6>
The similar document information output unit 6 outputs information about a document with high similarity calculated by the similarity calculation unit 5 to the display unit 7 (step S6).

  The similar document information output unit 6 may output information about a document whose similarity is higher than a predetermined threshold to the display unit 7, or display information about a predetermined number of documents from the higher similarity 7 may be output.

  The predetermined threshold and the predetermined number are appropriately determined according to required specifications. For example, when a picture book cover list is displayed on the display as shown in FIG. 12, a value that keeps the list while displaying the title in a size that can be read may be set in advance.

<Display unit 7>
The display unit 7 displays information about a document with high similarity output by the similar document information output unit 6 as a similar document search result (step S7). For example, the document title is displayed as a similar document search result. The similar document search result may include not only the document title but also information related to the document such as the attribute frequency series information of the document.

  Hereinafter, an example of a similar document search result displayed on the display unit 7 will be described.

  In FIG. 12, fifteen documents having high similarity to the attribute frequency series information of the document titled “Kairin Kirin” are displayed.

  In the display of FIG. 12, when the user rolls over or selects a document with high similarity with the pointing device, the attribute frequency series information of the document with high similarity is displayed as a similar document search result as shown in FIG. Also good.

In FIG. 13, since the document “5 hikininu” is rolled over or selected, the attribute frequency series information of the document titled “5 hikinuinu” is displayed in a pop-up.
Further, in FIG. 13, the attribute frequency series information of the document titled “A poor giraffe” as a search query is further displayed. In this way, the user can grasp how similar the attribute frequency series information of two documents is by further displaying the attribute frequency series information of the document as a search query as a similar document search result. It is possible to intuitively grasp the similarity of attribute frequency series information, that is, the similarity as a story due to, for example, the transition of “positive” and “negative” degrees (progress of story).

  In this example, the attribute frequency series information of a document titled “Kairin Kirin” as a query and the attribute frequency series information of a document titled “5 hikinuinu” as a search result are indicated by a solid line. There are parts where “positive” exceeds “negative” indicated by a broken line, but in most parts, “negative” exceeds “positive”, and “negative” exceeds “positive”. Are similar. In other words, they are similar in that they represent a sad story, both overall and ultimately. In this way, the document with the title “poor giraffe”, which is a query, is a sad story in which the number of words with the “negative” attribute exceeds the number of words with the “positive” attribute overall and ultimately. Similarly, the attribute frequency series information is similar to “poor giraffe” in the sense that the number of words with the “negative” attribute exceeds the number of words with the “positive” attribute, both overall and ultimately. However, it is determined that the document is similar to the document titled “5 Hiki no Inu” in the search result, which is also a sad story.

  Here, the attribute frequency series information of a certain document represents the story development of the document. For this reason, documents with similar attribute frequency series information are highly likely to be documents with similar stories. For this reason, a document with a similar story can be found by finding a document with similar attribute frequency series information. Thereby, it is possible to search for a document such as a picture book based on the type of story.

  In this case, the user may be able to adjust the curve of the attribute frequency series information of the document with the title “poor giraffe” that is displayed as a search query. In this case, the similar document search result corresponding to the adjusted attribute frequency series information may be displayed in real time.

[Modification]
<Modification 1>
The attribute frequency series information of each document corresponding to each of a plurality of different predetermined units is generated by the word extraction unit 1 and the attribute frequency series information generation unit 3, and is associated with the predetermined unit to be an attribute frequency series information storage unit 4 may be stored. For example, the attribute frequency series information of each document when the predetermined unit is one page and the attribute frequency series information of each document when the predetermined unit is two pages are the word extraction unit 1 and the attribute frequency series information. It may be generated by the generation unit 3 and stored in the attribute frequency series information storage unit 4.

  Note that the attribute frequency sequence information of each document when the minimum unit constituting the document is a predetermined unit is generated by the word extraction unit 1 and the attribute frequency sequence information generation unit 3, and is associated with the predetermined unit. It may be stored in the attribute frequency series information storage unit 4. In this case, it is assumed that information indicating the relationship between each of a plurality of different units and the minimum unit is also stored in the attribute frequency series information storage unit 4. At this time, when a predetermined unit is set by a granularity setting unit 71 described later, each of the units corresponding to the set predetermined unit is based on the relationship between the set predetermined unit and the minimum unit. The attribute frequency series information of the document is calculated and output to the similarity calculation unit 5.

  For example, the attribute frequency series information of each document having a page as a minimum unit is stored in the attribute frequency series information storage unit 4 in association with a predetermined unit. Furthermore, with respect to pages that are the smallest unit and chapters that are larger than the page, the correspondence between each chapter and the page (1 chapter is composed of 10 pages, 2 chapters are composed of 12 pages, etc.) is attribute frequency series information. It is stored in the storage unit 4. When the predetermined unit is set to “chapter” by the granularity setting unit 71, the attribute frequency series from the 1st to the 10th page based on the relationship between the predetermined unit “chapter” and the minimum unit “page” The attribute frequency series information of “Chapter 1” is calculated from the information. Similarly, the attribute frequency series information of “Chapter 2” is calculated from the attribute frequency series information from the 11th page to the 22nd page, and output to the similarity calculation unit 5 respectively. Is done. Note that the relationship between the predetermined unit and the minimum unit is not limited to the relationship between the chapter and the page. For example, when the sentence is the minimum unit, the predetermined unit may be a page or a chapter. That is, the relationship between the predetermined unit and the minimum unit may be a relationship between a page and a sentence or a relationship between a chapter and a sentence.

  In this case, the display unit 7 may display a granularity setting unit 71 capable of setting a predetermined unit. Examples of the granularity setting unit 71 are a button, a slider bar, and a pull-down menu, which will be described with reference to FIGS.

  The buttons are “Rough” and “Details” in “Picture books with similar story development” in FIG. Assume that the “large” button is associated with a unit larger than the unit associated with the “detail” button.

  When the user clicks the “Rough” button with the pointing device, the unit of the size associated with the “Rough” button in the attribute frequency sequence information stored in the attribute frequency sequence information storage unit 4 is selected. Based on the attribute frequency series information as a predetermined unit, the similarity calculation unit 5, the similar document information output unit 6, and the display unit 7 are processed. For example, the “Okaka” button is associated with a unit capable of performing a document search focusing on the outline of the story based on the average length of documents belonging to the document set to be searched. By performing a search by clicking the button, it is possible to perform a document search focusing on the outline of the story.

  Further, when the user clicks the “detail” button with the pointing device, the size corresponding to the “detail” button in the attribute frequency sequence information stored in the attribute frequency sequence information storage unit 4. Based on the attribute frequency series information whose unit is a predetermined unit, the similarity calculation unit 5, the similar document information output unit 6 and the display unit 7 are processed. By performing a search by clicking the “detail” button, it is possible to perform a document search focusing on the details of the story.

  As shown in FIG. 15, a slider bar capable of setting a predetermined unit and a “search” button may be displayed in the “picture book with similar story development”. A predetermined unit can be set by moving the slider of the slider bar with a pointing device. The slider bar slider may be continuously movable, or may be discontinuously movable only to a predetermined plurality of positions. The predetermined plurality of positions correspond to a plurality of units into which the document can be divided. The predetermined plurality of positions may correspond to different predetermined units corresponding to the attribute frequency series information stored in the attribute frequency series information storage unit 4.

  When the “search” button is clicked with the slider bar at the desired position, a unit corresponding to the position in the attribute frequency series information stored in the attribute frequency series information storage unit 4 is determined in advance. Based on the attribute frequency series information in the unit, the similarity calculation unit 5, the similar document information output unit 6 and the display unit 7 are processed.

  In the example of FIG. 15, buttons capable of setting a predetermined unit to a predetermined size are displayed at both ends of the slider bar. Assume that a unit larger than the unit associated with the “more detailed” button is associated with the “more rough” button.

  When the user clicks the “more rough” button with the pointing device, the first predetermined unit (for example, the unit set to the size corresponding to the “more rough” button) is set (for example, The attribute frequency stored in the attribute frequency sequence information storage unit 4 is a unit larger than the first predetermined unit) corresponding to the position at which the slider bar is placed at that time, as the second predetermined unit. Based on the attribute frequency sequence information corresponding to the second predetermined unit in the sequence information, the similarity calculation unit 5, the similar document information output unit 6 and the display unit 7 are processed. When the “more rough” button is clicked, the slider of the slide bar may be automatically moved to a position corresponding to the second predetermined unit.

  Further, when the user clicks the “more details” button with the pointing device, the third predetermined unit set at that time (the unit of the size associated with the “more details” button ( For example, a unit smaller than the third predetermined unit corresponding to the position where the knob of the slider bar is placed at that time is defined as a fourth predetermined unit, and is stored in the attribute frequency series information storage unit 4. Based on the attribute frequency series information corresponding to the fourth predetermined unit in the attribute frequency series information, the similarity calculation unit 5, the similar document information output unit 6, and the display unit 7 are processed. When the “more details” button is clicked, the slider of the slide bar may be automatically moved to a position corresponding to the fourth predetermined unit.

Further, as shown in FIG. 16, there is a pull-down menu in “a picture book with similar story development”, and a desired unit may be set by this pull-down menu. In the example of FIG. 16, items “Rough” and “Details” are expanded from the pull-down menu.
It is assumed that a unit larger than the unit size associated with the “detail” item is associated with the item “Roughly”.

  When the user selects the item “Rough” from the pull-down menu using the pointing device, the item is associated with the “Rough” item in the attribute frequency sequence information stored in the attribute frequency sequence information storage unit 4. Based on the attribute frequency series information in which a unit of a certain size is a predetermined unit, processing of the similarity calculation unit 5, the similar document information output unit 6, and the display unit 7 is performed.

  Further, when the user selects the “detail” item from the pull-down menu using the pointing device, the item is associated with the “detail” item in the attribute frequency sequence information stored in the attribute frequency sequence information storage unit 4. Based on the attribute frequency series information whose unit of size is a predetermined unit, processing of the similarity calculation unit 5, the similar document information output unit 6, and the display unit 7 is performed.

  As described with reference to FIGS. 14 to 16, the user may be able to set a predetermined unit using the granularity setting unit 71 before the search.

  On the other hand, as will be described below with reference to FIGS. 9, 17 and 18, before the search, the user cannot set a predetermined unit using the granularity setting unit 71, and a predetermined unit is set as the predetermined unit. Although the search is performed, the user may be able to set a predetermined unit using the granularity setting unit 71 displayed on the similar document search result display screen after the search.

  As shown in FIG. 9, there is no need for the buttons “Roughly” and “Details” in “Picture books with similar story development”. In this case, when the “story book with similar story development” button is clicked, an attribute having a predetermined unit as a predetermined unit in the attribute frequency series information stored in the attribute frequency series information storage unit 4 is selected. Based on the frequency series information, processing of the similarity calculation unit 5, the similar document information output unit 6, and the display unit 7 is performed.

  In this case, as shown in FIG. 17, a slider bar capable of setting a predetermined unit may be displayed on the display screen of the similar document search result of the display unit 7. The user can set a predetermined unit by moving the slider bar slider with a pointing device. The slider bar slider may be continuously movable, or may be discontinuously movable only to a predetermined plurality of positions. The predetermined plurality of positions correspond to a plurality of units into which the document can be divided. The predetermined plurality of positions may respectively correspond to different predetermined units corresponding to the attribute frequency series information stored in the attribute frequency series information storage unit 4. It is assumed that the initial position of the slider bar knob corresponds to the predetermined unit used when obtaining the similar document search result on the display unit 7, in other words, the initial value of the predetermined unit.

  In the example of FIG. 17, buttons capable of setting a predetermined unit to a predetermined size are displayed at both ends of the slider bar. Assume that a unit larger than the unit associated with the “more detailed” button is associated with the “more rough” button. In addition, a unit larger than the initial value of the predetermined unit is associated with the “more rough” button, and a unit smaller than the initial value of the predetermined unit is associated with the “more detailed” button. Suppose that As described above, at least one button associated with units having different sizes may be displayed on the display unit 7.

  When a predetermined unit is set by a slider bar or button, a similar document search result corresponding to the predetermined unit after setting may be displayed in real time. That is, the predetermined unit before setting by the granularity setting unit 71 is set as a first predetermined unit, the predetermined unit after setting by the granularity setting unit 71 is set as a second predetermined unit, and the display unit 7 The first similar document search result based on the first predetermined unit before setting by 71 and the second similar document search result based on the second predetermined unit after setting by the granularity setting unit 71 are displayed. May be. Thereby, the user can confirm whether the predetermined unit after the setting by the granularity setting unit 71 is appropriate.

  In the display of FIG. 17, when the user rolls over or selects one of the documents displayed as a document having high similarity to the query on the pointing device, the attribute frequency series information of the document having high similarity. Is displayed as a similar document search result. In the display of FIG. 17, the attribute frequency series information of the document that is the search query is further displayed as a similar document search result.

  As shown in FIG. 18, the predetermined unit before setting by the particle size setting unit 71 is set as the first predetermined unit, and the predetermined unit after setting by the particle size setting unit 71 is set as the second predetermined unit. The unit 7 may display the attribute frequency series information of the document serving as a search query based on the first predetermined unit and the attribute frequency series information of the document serving as a search query based on the second predetermined unit side by side. . In FIG. 18, the attribute frequency series information based on the first predetermined unit of the document that is the search query is displayed on the upper left, and the attribute frequency series information based on the second predetermined unit of the same document is displayed on the lower left. ing.

  The similar document search result based on the first predetermined unit before setting by the granularity setting unit 71 is set as the first similar document search result, and the similar document search result based on the second predetermined unit after setting by the granularity setting unit 71 If the user rolls over or selects a document with high similarity included in the second similar document search result with the pointing device, the attribute frequency of the selected document with high similarity Series information may be displayed. Since the user can simultaneously grasp the attribute frequency series information of the document serving as a search query based on different predetermined units before and after setting the predetermined unit, is the predetermined unit after setting appropriate? It can be confirmed more easily.

  It should be noted that the display of the similar document search result corresponding to the predetermined unit after the setting may be performed a plurality of times. That is, every time a predetermined unit is set by the slider bar or button, the similar document search result corresponding to the predetermined unit after setting is displayed in real time together with the similar document search result before setting by the granularity setting unit 71. May be.

  In FIG. 17 and FIG. 18, instead of the granularity setting unit 71 configured with a slider bar and buttons, only a button or a granularity setting unit 71 configured with a pull-down menu is provided. Alternatively, the user may be able to set a predetermined unit with the granularity setting unit 71 configured by a pull-down menu.

  As described with reference to FIGS. 14 to 16, even when the user can set a predetermined unit using the granularity setting unit 71 before the search, the description has been made with reference to FIGS. 17 and 18. The user may be able to set a predetermined unit using the granularity setting unit 71 displayed on the similar document search result display screen by the same method.

<Modification 2>
It is assumed that the attribute frequency series information is a histogram representing a series of attribute frequencies corresponding to words included in each of a plurality of segments constituting the document, and two attribute frequency series information for calculating similarity between each other Histograms, which are two pieces of attribute frequency series information for calculating the similarity between each other, are normalized so that the number of histogram bins becomes a predetermined number of bins, and the above 2 based on the normalized hisgram When the similarity is calculated, the granularity setting unit 71 may be able to set a predetermined number of bins.

  Examples of the granularity setting unit 71 are a button, a slider bar, and a pull-down menu, which will be described with reference to FIGS.

  For example, it is assumed that the “Rough” button in FIG. 14 is associated with a smaller number than the number associated with the “Detail” button.

  When the user clicks the “Rough” button with the pointing device, the similarity calculation unit 5 performs the process with the number of bins associated with the “Rough” button as the predetermined number of bins. That is, the granularity setting unit 71 outputs the number of bins associated with the “Rough” button to the similarity calculation unit 5 as the predetermined number of bins, and the similarity calculation unit 5 Based on the number of bins, the process of step S5 described above is performed. For example, the “Okaka” button is associated with the number of bins that can be used to perform a document search focusing on the outline of the story based on the average length of documents belonging to the document set to be searched. By performing a search by clicking the “Rough” button, a document search focusing on the outline of the story can be performed.

  Further, when the user clicks the “detail” button with the pointing device, the similarity calculation unit 5 performs processing with the number of bins associated with the “detail” button as the predetermined number of bins. That is, the granularity setting unit 71 outputs the number of bins associated with the “detail” button to the similarity calculation unit 5 as the predetermined number of bins, and the similarity calculation unit 5 Based on the number of bins, the process of step S5 described above is performed. By performing a search by clicking the “detail” button, it is possible to perform a document search focusing on the details of the story.

  As shown in FIG. 15, a slider bar that can set a predetermined number of bins and a “search” button may be displayed in “picture books with similar story development”. The number of bins can be set by moving the slider bar slider with a pointing device. The slider bar slider may be continuously movable, or may be discontinuously movable only to a predetermined plurality of positions. The predetermined plurality of positions correspond to the predetermined number of bins.

  When the “search” button is clicked with the knob of the slider bar placed at a desired position, the similarity calculation unit 5 process is performed with the number of bins corresponding to the position of the knob as the predetermined number of bins. That is, the granularity setting unit 71 outputs the number of bins associated with the position of the knob as the predetermined number of bins to the similarity calculation unit 5, and the similarity calculation unit 5 The above-described processing of step S5 is performed based on the number of.

  In the example of FIG. 15, buttons capable of setting a predetermined number of bins to a predetermined size are displayed at both ends of the slider bar. Assume that the “more rough” button is associated with a number of bins smaller than the number of bins associated with the “more detailed” button.

  When the user clicks the “more rough” button with the pointing device, the first predetermined bin set at that time is the same as the number of bins of the size associated with the “more rough” button. The similarity calculation unit uses the number of bins smaller than the number of bins (for example, the number of first predetermined bins corresponding to the position at which the slider bar is currently located) as the second predetermined number of bins. The similarity calculation unit 5 performs the process of step S5 described above based on the input number of predetermined bins. When the “more rough” button is clicked, the slider of the slide bar may be automatically moved to a position corresponding to the second predetermined number of bins.

  Further, when the user clicks the “more details” button with the pointing device, the third predetermined number set at that time is the same as the number of bins having the size associated with the “more details” button. Similarity calculation with the number of bins larger than the number of bins (for example, the third predetermined number of bins corresponding to the position where the slider bar is currently positioned) as the fourth predetermined number of bins The similarity calculation unit 5 performs the process of step S5 described above based on the input number of predetermined bins. When the “more details” button is clicked, the slider of the slide bar may be automatically moved to a position corresponding to the number of the fourth predetermined bin.

  Also, as shown in FIG. 16, there is a pull-down menu in “picture book with similar story development”, and the desired number of bins may be set by this pull-down menu. In the example of FIG. 16, items “Rough” and “Details” are expanded from the pull-down menu. It is assumed that the number of bins smaller than the number of bins associated with the “detail” item is associated with the item “Roughly”.

  When the user selects the item “Rough” from the pull-down menu by using the pointing device, the similarity calculation unit 5 performs processing with the number of bins associated with the “Rough” item as the predetermined number of bins. Is called. That is, the granularity setting unit 71 outputs the number of bins associated with the item “Rough” to the similarity calculation unit 5 as the number of predetermined bins, and the similarity calculation unit 5 Based on the number of bins, the process of step S5 described above is performed.

  Further, when the user selects the “detail” item from the pull-down menu by using the pointing device, the similarity calculation unit 5 performs processing using the number of bins associated with the “detail” item as the predetermined number of bins. Is done. That is, the granularity setting unit 71 outputs the number associated with the item “details” to the similarity calculation unit 5 as the number of predetermined bins, and the similarity calculation unit 5 The process of step S5 described above is performed based on the number.

  As described with reference to FIGS. 14 to 16, the user may be able to set a predetermined number of bins using the granularity setting unit 71 before the search.

  On the other hand, as will be described below with reference to FIGS. 9 and 17, the user cannot set a predetermined number of bins using the granularity setting unit 71 before the search, but sets a predetermined number of bins before the search. However, after the search, the user may be able to set a predetermined number of bins using the granularity setting unit 71 displayed on the similar document search result display screen.

  As shown in FIG. 9, there is no need for the buttons “Roughly” and “Details” in “Picture books with similar story development”. In this case, when the button of “picture book with similar story development” is clicked, the similarity calculation unit 5 performs processing with a predetermined number of bins as a predetermined number of bins. That is, the granularity setting unit 71 outputs a predetermined number of bins as a predetermined number of bins to the similarity calculation unit 5, and the similarity calculation unit 5 performs the above operation based on the input predetermined number of bins. The process of step S5 described is performed.

  In this case, as shown in FIG. 17, a slider bar capable of setting a predetermined number of bins may be displayed on the similar document search result display screen of the display unit 7. The user can set a predetermined number of bins by moving the knob of the slider bar with a pointing device. The slider bar slider may be continuously movable, or may be discontinuously movable only to a predetermined plurality of positions. The predetermined plurality of positions correspond to the predetermined number of bins. It is assumed that the initial position of the slider bar knob corresponds to the predetermined number of bins used when obtaining the similar document search result of the display unit 7, in other words, the initial value of the predetermined number of bins.

  In the example of FIG. 17, buttons that can set the number of predetermined bins to a predetermined number are displayed at both ends of the slider bar. Assume that the “more rough” button is associated with a smaller number of bins than the number of bins associated with the “more detailed” button. Further, the “more rough” button is associated with a smaller number of bins than the initial value of the predetermined number of bins, and the “more detailed” button is larger than the initial value of the predetermined number of bins. Assume that the number of bins is associated. Thus, at least one button associated with the number of bins having different sizes may be displayed on the display unit 7.

  When the predetermined number of bins is set by the slider bar or button, the similar document search result corresponding to the predetermined number of bins after the setting may be displayed in real time. That is, the number of predetermined bins before setting by the granularity setting unit 71 is the number of first predetermined bins, and the number of predetermined bins after setting by the granularity setting unit 71 is the number of second predetermined bins. The display unit 7 displays the first similar document search result based on the number of first predetermined bins before the setting by the granularity setting unit 71 and the second predetermined bin after the setting by the granularity setting unit 71 A second similar document search result based on the number may be displayed. Thereby, the user can confirm whether the number of the predetermined bins after the setting by the granularity setting unit 71 is appropriate.

  In the display of FIG. 17, when the user rolls over or selects one of the documents displayed as a document having high similarity to the query on the pointing device, the attribute frequency series of the document having high similarity. Information is displayed as a similar document search result. In the display of FIG. 17, the attribute frequency series information of the document that is the search query is further displayed as a similar document search result.

  The number of predetermined bins before setting by the granularity setting unit 71 is the number of first predetermined bins, and the number of predetermined bins after setting by the particle size setting unit 71 is the number of second predetermined bins. The display unit 7 includes a histogram that is attribute frequency series information of a document that is a search query normalized by the number of first predetermined bins, and a search query normalized by the number of second predetermined bins. A histogram that is attribute frequency series information of the document to be displayed may be displayed side by side. For example, a histogram that is attribute frequency series information normalized by the number of first predetermined bins of the document that is the search query is displayed on the upper left of the display screen of the display unit 7, and the lower left of the display screen of the display unit 7 A histogram which is attribute frequency series information normalized by the number of second predetermined bins of the same document may be displayed.

  The similar document search result based on the number of first predetermined bins before setting by the granularity setting unit 71 is set as the first similar document search result, and the similar document based on the number of second bins after setting by the granularity setting unit 71 When the search result is the second similar document search result, when the user rolls over or selects a document with high similarity included in the second similar document search result with the pointing device, the selected high similarity document is selected. Attribute frequency series information may be displayed. Since the user can simultaneously grasp the histogram that is the attribute frequency series information of the document that is the search query based on the number of different bins, it is further determined whether the predetermined number of bins after setting is appropriate. It can be easily confirmed.

  Note that the display of the similar document search result corresponding to the predetermined number of bins after the setting may be performed a plurality of times. That is, every time a predetermined number of bins is set by the slider bar or button, the similar document search result corresponding to the predetermined number of bins after setting is displayed together with the similar document search result before setting by the granularity setting unit 71. It may be displayed in real time.

  In FIG. 17, instead of the granularity setting unit 71 composed of a slider bar and a button, only a button or a granularity setting unit 71 composed of a pull-down menu is provided. The user may be able to set a predetermined number of bins with a granularity setting unit 71 configured with a menu.

  As described with reference to FIGS. 14 to 16, even when the user can set a predetermined number of bins using the granularity setting unit 71 before the search, it has been described with reference to FIG. 17. The user may be able to set a predetermined number of bins using the granularity setting unit 71 displayed on the similar document search result display screen by a similar method.

<Modification 3>
Note that the granularity setting unit 71 may be capable of setting both a predetermined unit and a predetermined number of bins.

  For example, S predetermined units are i (1),..., I (S), and S predetermined bins are j (1),..., J (S). Assume that a set of bin numbers (i (s), j (s)) (s = 1,..., S) is predetermined.

  The buttons, slider bar, and pull-down menu described with reference to FIGS. 14 to 18 include a predetermined unit and a predetermined number of bins (i (s), j (s)). (s = 1,..., S) are associated with each other, and the user operates a button, slider bar, and pull-down menu with a pointing device to set a predetermined unit and a predetermined number of bins (i (s ), j (s)) may be selectable.

  When the user selects a set of a predetermined unit and a predetermined number of bins (i (s), j (s)) by the user, the granularity setting unit 71 and the selected predetermined unit i (s) Information about the number of bins j (s) is output to the similarity calculation unit 5. The similarity calculation unit 5 includes attribute frequency sequence information having a predetermined unit i (s) selected among the attribute frequency sequence information stored in the attribute frequency sequence information storage unit 4 as a predetermined unit, The process of step S5 is performed based on the number of bins j (s).

  Summarizing Modifications 1 to 3, it can be said that the granularity setting unit 71 can set at least one of a predetermined unit and a predetermined number of bins.

<Modification 4>
For example, the granularity setting unit 71 increases as the predetermined unit has a smaller value or a predetermined number of bins as the change rate of the slope of the curve representing the attribute frequency series information depicted by the user by the method described with reference to FIG. 11 increases. A value may be set.

  For example, it is assumed that the curve representing the attribute frequency series information drawn by the user is the curve of FIG. In FIG. 19, the horizontal axis represents the page number, and the vertical axis represents a score that has a higher positive degree as it goes up and a negative degree as it goes down. The mountain valley with the steepest slope in the curve representing the attribute frequency series information drawn by such a user should be more meaningful than drawn by the user, so according to the steepness of the slope of the mountain valley, A predetermined unit or a predetermined number of bins is set. As a result, it is considered that a search suitable for the user needs can be performed.

  For example, the distance from the start point to the end point of the curve drawn by the user is X, the smallest distance from the maximum value to the minimum value of the curve drawn by the user is Δx, and the value below Δx is Δx ′. The particle size setting unit 71 sets the value obtained by rounding up X / Δx ′ as the number of bins, for example.

  Further, assuming that the average value of the total number S of segments is calculated for each different predetermined unit, the granularity setting unit 71 corresponds to the average value of the total number S of segments that is closest to the value obtained by rounding up X / Δx ′. The predetermined unit may be a predetermined unit.

  As described above, the granularity setting unit 71 sets a predetermined unit or a predetermined number of bins with such a fineness that can reflect a mountain and valley having the steepest slope in the curve representing the attribute frequency series information drawn by the user. Also good.

  The granularity setting unit 71 outputs information about the set predetermined unit and the predetermined number of bins to the similarity calculation unit 5. The similarity calculation unit 5 includes the attribute frequency series information corresponding to the set predetermined unit in the attribute frequency series information stored in the attribute frequency series information storage unit 4, and the set number of predetermined bins. Based on the above, the process of step S5 is performed.

<Modification 5>
The granularity setting unit 71 may determine a predetermined unit and a predetermined number of bins according to a document to be a query. For example, it is assumed that a predetermined unit and a predetermined number of bins are determined for each document serving as a query. For example, it is assumed that a predetermined unit and a predetermined number of bins are determined for each category such as a picture book, a children's book, and a document file to which a document to be queried belongs.

  When a query document is input, the granularity setting unit 71 reads a predetermined unit and a predetermined number of bins corresponding to the input query document, and reads the read predetermined unit and the predetermined number of bins. Is output to the similarity calculation unit 5. The similarity calculation unit 5 includes the attribute frequency series information corresponding to the set predetermined unit in the attribute frequency series information stored in the attribute frequency series information storage unit 4, and the set number of predetermined bins. Based on the above, the process of step S5 is performed.

<Modification 6>
The predetermined unit and the predetermined number of bins may be manually set by operating the granularity setting unit 71 when the user inputs a document to be a query.

  When the predetermined unit and the predetermined number of bins are set, the granularity setting unit 71 outputs information about the set predetermined unit and the predetermined number of bins to the similarity calculation unit 5. The similarity calculation unit 5 includes attribute frequency series information corresponding to a predetermined unit set in the attribute frequency series information stored in the attribute frequency series information storage unit 4, a set number of predetermined bins, The process of step S5 is performed based on the above.

<Modification 7>
A first word attribute assignment in which an attribute assignment corresponding to each word is defined, and a second word attribute assignment in which an attribute assignment corresponding to each word is defined, which is different from the first word attribute assignment. Suppose that there is.

  In this case, the similar document search device may perform the search based on the second word attribute assignment after filtering the document by the search based on the first word attribute assignment. For example, the first word attribute assignment and the second word attribute assignment are set so that the type of the first word attribute assignment and the type of the second word attribute assignment are different. For example, the first word attribute assignment is assigned using the polarities of “joy”, “anger”, “sorrow”, and “easy”, and the second word attribute assignment is the topic extraction using the topic model described in Reference 5. Allocation using the method of Thereby, search results can be narrowed down from different viewpoints.

  That is, assuming that the attribute of each word based on the first word attribute assignment and the attribute of each word based on the second word attribute assignment are stored in the word attribute storage unit 2, the similar document search device A temporary similar document is obtained by performing processing of the attribute frequency series information generation unit 3, the similarity calculation unit 5, and the similar document information output unit 6 based on the attribute of the word based on the first word attribute assignment read from the storage unit 2. After obtaining the temporary similarity document, the attribute frequency series information generation unit 3, the similarity calculation unit 5, and the like based on the word attributes based on the second word attribute assignment read from the word attribute storage unit 2 A final similar document search result may be obtained by performing processing of the similar document information output unit 6.

  The user may be able to select each of the first word attribute assignment and the second word attribute assignment. As a result, it is possible to perform a search in which the first word attribute assignment is more important than the second word attribute assignment. In addition, when the number of documents presented by the search by the first word attribute assignment is large, it is possible to narrow down from the viewpoint of the second word attribute assignment by performing the search by the second word attribute assignment. it can. Thereby, the user can perform a more desired search.

<Modification 8>
The processing of the word extraction unit 1 and the attribute frequency series information generation unit 3 may be performed prior to the processing of the similarity calculation unit 5, the similar document information output unit 6, and the display unit 7, or the word extraction unit 1 and the attribute frequency The processing of the sequence information generation unit 3 and the processing of the similarity calculation unit 5, the similar document information output unit 6, and the display unit 7 may be performed in parallel.

  The processing from step S1 to step S3 performed using the word extraction unit 1, the word attribute storage unit 2, and the attribute frequency series information generation unit 3 is preprocessing prior to similarity calculation and similar document search. For this reason, the similarity calculation device 20 and the similar document search device may not include the word extraction unit 1, the word attribute storage unit 2, and the attribute frequency series information generation unit 3 surrounded by a dashed line in FIG. Further, the similarity calculation method and the similar document search method may not include the processing of step S1 and step S3 surrounded by a one-dot chain line in FIG.

  In other words, the similarity calculation device 20 may include only the attribute frequency series information storage unit 4 and the similarity calculation unit 5. Further, the similar document search apparatus may include only the attribute frequency series information storage unit 4, the similarity calculation unit 5, the similar document information output unit 6, and the display unit 7. Moreover, the similarity calculation method may have only the process of step S5. Further, the similar document search method may include only the processing from step S5 to step S7.

  In these cases, the attribute frequency series information storage unit 4 includes a word extraction unit 1, a word attribute storage unit 2, and an attribute frequency series information generation unit 3 provided in a device different from the similarity calculation device 20 and the similar document search device. Assume that the attribute frequency series information generated by is stored in advance.

  Needless to say, other modifications are possible without departing from the spirit of the present invention.

<Program and recording medium>
When the processing in each part of the similar document search apparatus or similarity calculation apparatus 20 is realized by a computer, the processing contents of the functions that each part of the similar document search apparatus or similarity calculation apparatus 20 should have are described by a program. Then, by executing this program on a computer, the processing of each part is realized on the computer.

  The program describing the processing contents can be recorded on a computer-readable recording medium. As the computer-readable recording medium, for example, any recording medium such as a magnetic recording device, an optical disk, a magneto-optical recording medium, and a semiconductor memory may be used.

  The processing of each unit may be configured by executing a predetermined program on a computer, or at least a part of these processing may be realized by hardware.

DESCRIPTION OF SYMBOLS 1 Word extraction part 2 Word attribute storage part 3 Attribute frequency series information generation part 4 Attribute frequency series information storage part 5 Similarity calculation part 6 Similar document information output part 7 Display part 71 Granularity setting part 72 Input part 8 Attribute assignment part 20 Similarity Sex calculator

Claims (12)

A document is assumed to be composed of a plurality of segments obtained by dividing the document into predetermined units, and information indicating a frequency series of attributes corresponding to words included in each of the plurality of segments constituting the document is provided. As attribute frequency series information,
An input unit for receiving input of attribute frequency series information selected or drawn by the user;
A display unit that displays information about a document having attribute frequency series information having high similarity to the input attribute frequency series information as a similar document search result;
Similar document search device including The similar document search device according to claim 1,
A particle size setting unit capable of setting the predetermined unit;
Similar document search device. The similar document search device according to claim 1,
The attribute frequency series information is a histogram representing a frequency series of attributes corresponding to words included in each of a plurality of segments constituting the document, and two attribute frequency series information for calculating a similarity between each other Histograms that are two pieces of attribute frequency series information for calculating the degree of similarity of each other so that the number of bins of the histogram is the predetermined number of bins are normalized, and the above 2 based on the normalized hisgram As the similarity is calculated,
A particle size setting unit capable of setting the number of the predetermined bins;
Similar document search device. The similar document search device according to claim 2 or 3,
The granularity setting unit is at least one button or pull-down menu associated with a different unit or number of bins displayed on the display unit.
Similar document search device. The similar document search device according to claim 2 or 3,
The particle size setting unit is a slider bar displayed on the display unit.
Similar document search device. The similar document search device according to claim 5,
The granularity setting unit further includes at least one button or pull-down menu that is displayed on the display unit and associated with units of different sizes or the number of bins.
Similar document search device. The similar document search device according to claim 1,
When the attribute frequency series information received by the input unit is a curve, the smaller the change rate of the slope of the curve representing the attribute frequency series information, the larger the smaller value as the predetermined unit or the larger the predetermined number of bins. It further includes a granularity setting unit for setting a value.
Similar document search device. The similar document search device according to any one of claims 2 to 7,
The display unit displays the similar document search result before setting by the granularity setting unit, and displays the similar document search result after setting by the granularity setting unit.
Similar document search device. The similar document search device according to any one of claims 2 to 7,
The display unit displays the similar document search result before setting by the granularity setting unit, and the predetermined unit of the granularity setting unit or the predetermined unit by an operation of a user who viewed the similar document search result before setting. When the number of bins is set, the similar document search result after the setting is displayed.
Similar document search device. The similar document search device according to any one of claims 1 to 9,
The display unit includes attribute frequency series information of a document that the user has rolled over or selected with a pointing device in a document having attribute frequency series information having high similarity to the input attribute frequency series information, and the user The selected or drawn attribute frequency series information is further displayed as the similar document search result.
Similar document search device. A document is assumed to be composed of a plurality of segments obtained by dividing the document into predetermined units, and information indicating a frequency series of attributes corresponding to words included in each of the plurality of segments constituting the document is provided. As attribute frequency series information,
An input step for receiving an input of attribute frequency series information selected or drawn by the user;
A display step for displaying information about a document having attribute frequency series information having a high similarity to the input attribute frequency series information as a similar document search result;
Similar document search method including The program for functioning a computer as each part of the similar document search apparatus in any one of Claim 1 to 10.