JPH07325837A - Communication sentence retrieval device by abstract word and communication sentence retreival method by the abstract word - Google Patents

Abstract

PURPOSE:To provide a communication sentence retrieval device and a communication sentence retrieval method by abstract words capable of automatically retrieving a communication sentence by the retrieval keyword of fuzzy expression when an abstract keyword is inputted. CONSTITUTION:This device is provided with a similar word acquisition means 100 for extracting the abstract word from a retrieval request sentence and acquiring at least more than one similar words similar to the abstract word, an evaluation means 200 for retrieving the communication sentence example storage means of a retrieval object by the similar words obtained by the similar word acquisition means 100 and evaluating the probability of a retrieved communication sentence candidate and an output means 300 for outputting an evaluated result evaluated by the evaluation means 200 and the communication sentence candidate.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a communication sentence retrieval device using abstract words and a communication sentence retrieval method using abstract words, and in particular, in a Japanese communication sentence processing device or the like, an abstract noun or adjective / adjective verb of a user.・ Analysis of requests that include adverbs and conversion to keywords that correspond to adjectives, adjective verbs, adverbs, and abstract nouns, so that the communication text is ambiguous, but communication using abstract words that are more familiar to users The present invention relates to a sentence search device and a communication sentence search method using abstract words.

[0002]

2. Description of the Related Art In a conventional document retrieval apparatus, a physical / objective keyword is used for retrieval. For example, a document search device for searching newspaper articles uses nouns such as a person's name, learning terms, product name, etc. as a keyword used for the search. When using adjectives, adjectives pointing to physical parameters are used. For example, a "new" product, which is an adjective that identifies the most recently released product, i.e., the most recently released product in physical time. Similarly, the sentence "high price" indicates that the value is large in the physical dimension of prices. Such adjectives can be defined using "index of physical dimension (price, height, temperature) that can be described" and "value on physical dimension". This is because the physical dimensions are basically independent, and there is a clear scale for the physical quantity.

The words necessary for searching such newspapers are those expressed by concrete objects, clear conceptual nouns, and physical attributes, and the words used as search keywords are formally defined.

Also, when searching for a message such as a message, it is necessary to be able to process abstract keywords. A conventional message retrieval device uses a sender and a receiver as keys to retrieve, and a method of performing a retrieval by a politeness indicating the politeness of the contents of a communication and an intimacy indicating the familiarity (Japanese Patent Application No. 3-236191). Further, referring to the example sentence collection, there are other humor type / hiding type, heartwarming type, expecting type, etc. as indexes of communication sentences.

[0005]

However, in order to search for a message such as a message, the above-mentioned conventional document search device uses the adjectives and adjectives relating to human senses and feelings for each communication example sentence. There is also a method of assigning an index expressed by a verb / adverb, but this must perform the same evaluation for a new example sentence.

The present invention has been made in view of the above points, and solves the above problems of the related art, and when an abstract keyword is input, a communication message is automatically generated using an ambiguous search keyword. An object of the present invention is to provide a communication text search device and a communication text search method using abstract words that can be searched.

A further object of the present invention is to specify an adjective that modifies an ambiguous trait of a message or a noun indicating an abstract concept as a keyword. By converting from the attribute to a specific word string on the surface of the message and searching using the word string on the surface, ambiguous adjectives with meanings such as "warm", "heartwarming", and "hope" Adjective verb
An object of the present invention is to provide a communication sentence search device using an abstract word and a communication sentence search method using an abstract word that can search a communication sentence using an adverb or an abstract noun as a keyword.

[0008]

FIG. 1 is a block diagram showing the principle of the present invention.

A communication text search device using abstract words according to the present invention inputs a search request text for searching a communication text, searches a communication text example storage means for storing the communication text written in Japanese, and makes a search request. In a search device that searches a communication sentence corresponding to a sentence, a similar word acquisition unit 100 that extracts an abstract word from a search request sentence and acquires at least one similar word similar to the abstract word.
And an evaluation unit 200 that searches the communication sentence example storage unit that is the search target with the similar words obtained by the similar word acquisition unit 100 and evaluates the certainty of the retrieved communication sentence candidates, and the evaluation evaluated by the evaluation unit 200. It has a result and an output means 300 for outputting a message candidate.

Further, the above-mentioned similar word acquisition means 100 is
Morphological analysis means 2 for dividing the input search request sentence into morphemes, abstract word mutual meaning distance storage means 9 for storing the semantic distance between the abstract words, and abstract search from the search request sentence. Search keyword extracting means 3 for extracting keywords by referring to the abstract word mutual meaning distance storing means 9, and abstract word mutual meaning distance storing means 9 for abstract words having meanings close to the extracted abstract search keywords.
Search keyword conversion means 4 for searching by referring to, a communication sentence search word meaning distance storage means 10 storing a meaning distance between an abstract word and a word that actually appears in a communication sentence,
It includes a search keyword expanding means 5 for expanding the search keyword with words that are semantically similar to the search keyword obtained by the search keyword converting means 4 by referring to the communication sentence search word meaning distance storing means 10.

The above-mentioned abstract word mutual meaning distance storing means 9 has mutual similarity information of words having ambiguous meanings.

The communication sentence search word meaning distance storage means 10 has similarity information between the keyword used in the communication sentence and the abstract word obtained by the search keyword conversion means.

FIG. 2 illustrates the principle of the present invention.

Further, the abstract sentence communication message search method of the present invention inputs a search request sentence for searching a communication sentence (step 1), and accumulates communication sentences written in Japanese. To retrieve a communication sentence corresponding to the search request sentence, an abstract word is extracted from the search request sentence to obtain at least one similar word similar to the abstract word (step 2) Search the communication sentence example storage means to be searched with similar words (step 3), evaluate the certainty of the retrieved communication sentence candidates (step 4), and output the evaluation result and the communication sentence candidates (step 4). Step 5).

Further, in the above-mentioned abstract sentence correspondence search method, adjectives, adjective verbs, adverbs, and nouns having abstract meanings are prepared as search keywords, and adjectives, adjective verbs, and adverbs with ambiguous meanings are prepared. For nouns showing abstract meaning, set the mutual semantic distance for all words, refer to the mutual semantic distance, expand the keywords with words that are semantically similar, The expanded message keyword is used to search the message example storage unit, the likelihood of the keyword of the searched message is evaluated, and the searched message and the evaluation value of the message are output.

[0016]

The present invention is applicable to all words such as adjectives, adjective verbs, adverbs and nouns having abstract meanings such as "warm", "heartwarming" and "hope" as search keywords for communication sentences, for example, telegram sentences. A keyword for searching the message storage means is acquired using a table indicating the degree of similarity between each other, and the acquired keyword is expanded to obtain a plurality of similar keywords. Since the storage means is searched, a wide range of search is possible.

Further, by obtaining and outputting the certainty of the retrieved communication text, even if an ambiguous request text is input, various keywords can be generated and communication text candidates with high similarity can be searched. Also, it is possible to automatically retrieve and output the communication text having the highest degree of similarity.

[0018]

Embodiments of the present invention will be described below with reference to the drawings.

FIG. 3 shows the configuration of a message retrieval device according to an embodiment of the present invention. The communication text search device shown in FIG. 1 includes a user request input unit 1, a morpheme analysis unit 2, a search keyword extraction unit 3, a search keyword conversion unit 4, a search keyword expansion unit 5, a keyword search unit 6, and a search communication message evaluation unit 7. , A communication sentence output unit 8, an abstract word mutual meaning distance table 9, a communication sentence search word meaning distance table 10, and a communication sentence example storage device 11.

In the above communication text search device, when a character string expressing a search request from a user is input to the user request input unit 1, the morpheme is input to the morpheme analysis unit 2 with respect to the morpheme. Analyze, divide the input sentence into morphemes, and if the morphemes are verbs / adjectives / adjective verbs,
Create and output a list including the final form.

The search keyword extraction unit 3 extracts a word serving as a search keyword from the morpheme analysis result input from the morpheme analysis unit 2.

As shown in FIG. 4, the keyword mutual meaning distance table 9 includes n abstract words c _i (1 ≦ i ≦) that are adjectives, adjective verbs, adverbs, and abstract nouns that can be keywords.
(n: i is an integer) is stored in the form of an end form, and the end form of the input morpheme string and the abstract word c _i are compared with each other to list 0 or more matching abstract words c _i . Output in the format. For example, if the abstract words c ₂ and c ₉ match, the following list is output.

Output example: (c ₂ , c ₉ ) The search keyword conversion unit 4 refers to not only the abstract word input from the search keyword extraction unit 3 but also the abstract word mutual meaning distance table 9 to obtain similar meanings. Search for the abstract word and add it to the list.

In many cases, the individual abstract words input to the search keyword conversion unit 4 do not exist in the actual message. Therefore, the input abstract word is increased by the semantically close abstract word.

In the abstract word mutual meaning distance table 9, the semantic distance between n abstract words is set in advance. For example, it means the distance d _ij between abstract word c _i and c _j, if the meaning of abstract words c _i and c _j is identical, the "0", if the most far, so that "1" And 0 ≦ d _ij
It takes a value of ≦ 1. Here, the mutual semantic distance d _xy between the input x-th abstract word c _x and the y-th abstract word c _y is d _xy ≦ T _{a with} respect to a preset threshold T _a .
Cases, i.e., if semantically similar are abstract words c _y be in conjunction with the mean distance d _xy LOV (c
₅ d ₂₅ ) and output. In the output example, if d ₂₅ ≦ T _a and d ₂₈ ≦ T _a for the abstract word c ₂ , a new list of (c ₅ d ₂₅ ) (c ₈ d ₂₈ ) is added. The abstract word c _x that is the input
The value of the semantic distance to itself is “0”.

Output example: (c ₂ 9) (c ₉ 0) (c ₅
d ₂₅ ) (c ₈ d ₂₈ ). The search keyword expansion unit 5 uses (abstract word c _i meaning word d
A word for actually searching for a communication sentence in the communication sentence example storage device 11 by referring to the search word meaning distance table 10 in which words such as nouns and verbs used in the communication sentence are accumulated by inputting the list _i ) To the input list.

FIG. 5 shows a single message retrieval unit according to an embodiment of the present invention.
The structure of a word meaning distance table is shown. In the figure, communication
The sentence search word meaning distance table 10 has n abstract words c.
_i(1 ≦ i ≦ n: i is an integer) and k search words w_j(1
≦ j ≦ k: j is an integer) and the semantic distance t is stored.
It Abstract word c_iAnd the y-th search word w_jMeaning between
If they are the same, the semantic distance t_ij= 0, most distant
If t_ij0 ≦ t, such as = 1_ijA value of ≦ 1
take. Also at this time, the threshold T of the semantic distance is set in advance._bSet
Ku. Yes (c_xc_xy) Input list
Search word meaning distance table 10c_xColumn, search
Word w₁~ W_kSemantic distance t to_x1~ T _xkIs the value of t
_x1~ T_b, That is, the abstract word c_xA certain search word w
_zC have semantically close values, then c_xAnd w_zMeaning distance
Is t_xzTo extract. However, (c in the input list_xd
_xy) In c_xMeaning distance d_xyCorrection to also consider
Meaning distance after D_ZIs d_xyAnd t_xzIs calculated by the following formula
It

D_z= 1- (1-t_xz) × (1-d_xy) Is calculated. This meaning distance D_zIs d_xyWhen = 0, immediately
When the meaning distance is short, D_Z= T_xzNext, meaning distance
d_xyIs increased, D_zAlso increases. The value is finally checked.
0 or more at the end of the input list of the keyword expansion unit 5
Search word w above_zAnd meaning distance D_zOf (w_zD_z) Added
The output list is output. For example, c₂And w₂Means
Distance t_{twenty two}≤T_bIt is close in meaning and is an abstract word c₂And correspondence
Search word w ₆Is t₂₆≤T_b, Then (w₂D₂)
(W₆D₆) Is newly added to the list, and the output is
It becomes a street.

Output example: ((c ₂ 0) (c ₉ 0)
(C ₅ d ₂₅ ) (c ₈ d ₂₈ ) (w ₂ D ₂ ) (w
₆ D ₆ )) Here, in the output list, the search keyword extracted by the search keyword extraction unit 3 (c _n 0) and converted by the search keyword conversion unit 4 (c _x d _xy ), the search keyword Extension 5
The output of (w _z D _z ) expanded by 3 is connected, but basically all of them are (word meaning distance)
Is output as a list. Therefore, if there are q lists in total, the word end form W _i (1 ≦ i ≦ q:
i is an integer) and the semantic distance D _i is expressed as follows.

((W₁ D₁) (W₂D₂) (W₃
D₃)… (W_qD_q)) In the keyword search unit 6, (W_iD_i) List
Are processed one by one from the beginning. Communication example storage device 1
1 has p messages M_j(1 ≦ j ≦ p: j is an integer)
The verbs, adjectives, and adjectives that are morphologically analyzed and used are
The final form is also hurriedly accumulated. The keyword search unit 6
p message M_jOut of q (W _iD_i) Squirrel
Word W in_iSearch for correspondence that contains. Search
Correspondence M_jAnd the message M_jWas included in
Word W_iList of (W_iD_i) Is newly created. this
Such p messages M_jList of words that were present (W
_imD_im) Is output. ((M₁(W₁₁ D₁₁) (W₁₂ D₁₂) (W₁₃ 
D₁₃)) (M₂(W_{twenty one} D_{twenty one}) (W_{twenty two} D_{twenty two}) (W_{twenty three} D_{twenty three})
(W_{twenty four} D_{twenty four}) ・ ・ (M_p(W_j1 D_j1)) The search message evaluation unit 7 searches the r search message M searched for.
_jEvaluate the input list sequence one by one for the message M
_jConfidence S that indicates the certainty of search results_jCalculate
It Confidence S_jIs the message M_jThe word W that existed in_jm
Meaning distance D_jmUsing (1-D_jm) Value for all m
Is the sum of. Meaning distance D_jmSum of
The smaller the confidence S_jThe value of becomes large. For example, ((M₁(W₁₁ D₁₁) (W₁₂ D₁₂) (W₁₃ 
D₁₃)) From the list₁Confidence S₁Ask for
And S₁= (1-D₁₁) + (1-D₁₂) + (1-D₁₃).

Such a certainty factor is calculated for all communication sentences and output as a list.

((M ₁ S ₁ ) (M ₂ S ₂ ) ... (M _r S _r ) The message output unit 8 sorts by paying attention to the certainty factor S _j in the input list. The output is arranged in the list in descending order of the degree S _j .

Next, a specific example will be described.

First, the user inputs a request sentence "Is there a warm telegram?" To the user request input unit 1. This request is morphologically analyzed by the morphological analysis unit 2. For example, it is divided into morphemes as follows. "/" Is a morpheme delimiter.

Heartwarming / Yes / Telegram / Ha / Yes / No / No / In the above morphological analysis, when the morpheme is a verb / adjective / adjective verb, the final form information is also added.

The search keyword extraction unit 3 refers to the abstract word mutual meaning distance table 9 and refers to the n abstract words c _i (1
≦ i ≦ n: i is an integer), and c _i that matches the final form of the morphological analysis result is extracted. An example of the abstract word mutual meaning distance table 9 is shown in FIG. In this case, the adverb "heartwarming" matches, and the output is in the following list format.

The (heartwarming) search keyword conversion unit 4 uses the abstract word mutual meaning distance table.
Bull 9 to find other similar abstract words
The distance is also output. Threshold value T for extraction _a= 0.
Set to 3. The input abstract word itself has a semantic distance of 0
To create and output a list. The following list is shown in Figure 6.
With reference to the abstract word mutual meaning distance table 9
is there.

((Heartwarming 0) (warm 0.2)
(Nagoyaka 0.3) (Family 0.15)) The search keyword expansion unit 5 refers to the communication sentence search word meaning distance table 10 to calculate the actual communication message from the input abstract word c _x . Output the words used for the search. FIG. 7 shows an example of the configuration of the communication sentence search word meaning distance table. When the communication sentence search word meaning distance table 10 is referred to as the threshold value T _b = 0.3 for the input abstract word c _x ,
The following list string (search word meaning distance) is obtained. In FIG. 7, the vertical axis represents the message search word,
The horizontal axis represents the abstract word c _n (abstract noun / adjective / adjective verb / adverb). Therefore, for example, (heartwarming 0.0) → (lights 0.15) (dream 0.
3) (balloons 0.25) (soyokaze 0.2) (warm 0.2) → (spring 0.05) (fresh green
0.1) (Poka Poka 0.1) (Kotatsu 0.1) (Nagoyaka D 0.3) → (Sakura 0.2) (Koharu
0.1) (Shiosai 0.25) (Homely 0.15) → (Miso soup 0.1) (Chopsticks
0.15) (potato 0.2) Next, in order to consider the semantic distance d of the input abstract word,
When the semantic distance of the message search word is t, the final semantic distance D is as follows.

D = 1- (1-t) (1-d) Specifically, (light 0.15) is (heartwarming
0.0), D = 1- (1-0.15) (1-0) = 0.15 can be calculated, and the list becomes (lamp 0.15).

(Search word meaning distance) is (Spring 0.0
For (5), considering (warm 0.2), D = 1- (1-0.05) (1-0.2) = 0.24, and the list is set as (spring 0.24). It The output is the one in which the communication sentence search word list is connected to the end of the input list. The output list is shown below. ((Heartwarming 0) (warm 0.2) (nagoya 0.3) (homely 0.15) (light 0.1
5) (Fresh green 0.28) (Poka Poka 0.28) (Kotatsu 0.28) (Sakura 0.44) (Koharu 0.3
7) (Shiosai 0.48) (Miso soup 0.24) (Chopsticks
0.28) (potato 0.32)) The keyword search unit 6 receives the q (word W _i
Using the word W _i · in the list of the semantic distance D _i ) as a keyword, the communication text stored in the communication text example storage device is searched. For example, in the message M ₁ , (heartwarming 0) (balloon 0.25) (fresh green 0.
28), if there are three keywords, the keyword search unit 6
Outputs a list such as (M ₁ (heartwarming 0) (balloon 0.25) (fresh green 0.28)). The search is performed for all telegrams, and the following list is finally output.

((M ₁ (heartwarming 0) (balloon
0.25) (fresh green 0.28)) (M ₂ (fresh green 0.28) (soft 0.3) (heartwarming 0)) (M ₃ (warm warm 0.2) (Shiosai 0.48) (soyokaze 0.2) (Koharu 0.37))) The search message evaluation unit 7 evaluates the accuracy of the search by evaluating the semantic distance of the search word existing in the message M _j .
As an example of a method of calculating the evaluation value E, a method of using the sum of the values of (1-D _jm ) of the semantic distance D _jm of the search word will be shown below.

(M ₁ (Honoponoshita 0) (Balloons 0.
25) (fresh green 0.28)) Certainty factor S ₁ = (1-0) + (1-0.25) + (1-
0.28) = 2.47 Thus, the following list is created as the evaluation result of the message M ₁ .

(M ₁ 2.47) Similarly, the remaining communication messages M ₂ and M ₃ are evaluated and sorted by the evaluation value. The larger the evaluation value is, the more “faint” message can be retrieved.

((M ₁ 2.75) (M ₂ 2.47) (M
₃ 2.42)) In the case of this example, the message M ₁ is the most “heartwarming” message.

The message output unit 8 outputs the sorted messages M ₁ , M ₂ and M ₃ with information on the likelihood of retrieval being added.

As described above, the adjectives, adjective verbs, adverbs, and nouns having abstract meanings such as "warm", "heartwarming", and "hope" are used as keywords for telegram literature search, and ambiguous by using the message search words. Meaning adjectives, adjective verbs,
For an adverb and a noun indicating an abstract meaning, an abstract word mutual meaning distance table 9 in which mutual meaning distances are set for all abstract words is provided, and a correspondence sentence search word meaning distance table 10 is used. The meaning of the search word existing in the communication text is searched by expanding the keyword with a word that is semantically similar and searching the communication example storage device 11 with the expanded keyword, in addition to the condition that the search is simply matching the keyword. By evaluating the distance, the certainty of the search is compared numerically, and the retrieved communication message is output. As a result, the search is automatically realized using the adjectives that modify the ambiguous traits of the message and the nouns that represent abstract concepts as keywords.

The values of the meaning distances of the abstract word meaning distance table 9 of FIG. 6 and the communication sentence search word meaning distance table 10 of FIG. 7 used in the above embodiment are not limited to this example, but may be various values. Design is possible.

Further, the present invention is not limited to the above-mentioned embodiments, and various modifications and applications are possible within the scope of the claims.

[0049]

As described above, according to the present invention, by inputting a search request sentence using an adjective / adjective verb / adverb indicating an emotion / sense, or a noun indicating an abstract concept, Not only words but also words that are close in meaning are searched from the semantic distance table, the correspondence database is searched by keywords in consideration of the semantic distance, and correspondence messages that meet the search conditions are surely searched from the means for storing correspondence messages. It can be output including the likelihood.

Therefore, for example, it is possible to present a user with a plurality of communication sentences in the order of high evaluation, and to select a desired communication message. Furthermore, the communication message with the highest evaluation degree is automatically used. It is also possible to compose a telegram sentence or the like.

[Brief description of drawings]

FIG. 1 is a principle configuration diagram of the present invention.

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

FIG. 3 is a configuration diagram of a message search device according to an embodiment of the present invention.

FIG. 4 is a configuration diagram of an abstract word mutual meaning distance table according to an embodiment of the present invention.

FIG. 5 is a configuration diagram of a communication sentence search word meaning distance table according to an embodiment of the present invention.

FIG. 6 is a diagram showing a configuration example of an abstract word meaning distance table according to an embodiment of the present invention.

FIG. 7 is a diagram showing a configuration example of a communication sentence search word meaning distance table according to an embodiment of the present invention.

[Explanation of Codes] 1 user request input unit 2 morphological analysis unit 3 search keyword extraction unit 4 search keyword conversion unit 5 search keyword expansion unit 6 keyword search unit 7 search message evaluation unit 8 message output unit 9 abstract word meaning distance table 10 Communication sentence search word meaning distance table 11 Communication sentence example storage device 100 Similar word acquisition means 200 Evaluation means 300 Output means

Claims (6)

[Claims] 1. A search request sentence for searching a communication sentence is input, a communication sentence example accumulating means for accumulating a communication sentence written in Japanese is searched, and a communication sentence corresponding to the search request sentence is searched. In the search device, an abstract word is extracted from the search request sentence, and a similar word acquiring unit that acquires at least one similar word similar to the abstract word; An evaluation unit that searches the communication-sentence example storage unit to be searched with the similar word and evaluates the certainty of the retrieved communication-sentence candidate, an evaluation result evaluated by the evaluation unit, and an output that outputs the communication-sentence candidate. A communication sentence search device using an abstract word, which comprises: 2. The similar word acquisition means, a morpheme analysis means for dividing the input search request sentence into morphemes, an abstract word, and an abstract word mutual meaning distance storing a meaning distance between the abstract words. Storage means, search keyword extraction means for extracting an abstract search keyword from the search request sentence by referring to the abstract word mutual meaning distance storage means, and an abstract word having a meaning close to that of the extracted abstract search keyword Search keyword conversion means for searching the abstract word mutual meaning distance storage means, and a communication sentence search word meaning distance storage storing a meaning distance between the abstract word and a word that actually appears in the communication sentence And a search key with a word that is semantically similar to the search keyword obtained by the search keyword conversion means by referring to the communication sentence search word meaning distance storage means. The communication message search device using abstract words according to claim 1, further comprising search keyword expansion means for expanding a keyword. 3. The communication message search device using abstract words according to claim 2, wherein the abstract word mutual meaning distance storage means has similarity information between words having ambiguous meanings. 4. The abstract according to claim 2, wherein the correspondence sentence search word meaning distance storage unit has similarity information between a keyword used in the correspondence sentence and an abstract word obtained by the search keyword conversion unit. Communication sentence retrieval device by word. 5. A search request sentence for searching a communication sentence is input, a communication sentence example accumulating unit that accumulates the communication sentence written in Japanese is searched, and a communication sentence corresponding to the search request sentence is searched. In this case, an abstract word is extracted from the search request sentence, at least one or more similar words similar to the abstract word are acquired, and the communication sentence example accumulating means to be searched is searched with the similar word. A method for searching a communication message by an abstract word, characterized in that the certainty of the retrieved communication message candidate is evaluated and the evaluation result and the communication message candidate are output. 6. An adjective, an adjective verb, an adverb, or a noun having an abstract meaning is prepared as a keyword for a search, and an adjective, an adjective verb, or an adverb with an ambiguous meaning and a noun indicating an abstract meaning are provided. Mutual semantic distances are set for all words, the mutual semantic distances are referred to, the keywords are expanded with words that are semantically similar to each other, and the communication sentence example accumulation is performed with the expanded keywords. The method of searching a communication message by an abstract word according to claim 5, wherein the means for searching is searched for, the likelihood of the keyword of the searched communication message is evaluated, and the searched communication message and the evaluation value of the communication message are output.