JP3428554B2 - Semantic network automatic creation device and computer readable recording medium - Google Patents

Description

Detailed Description of the Invention

[0001]

The present invention relates to the Internet,
The present invention relates to a device for automatically creating a semantic network used for knowledge acquisition and knowledge processing in an electronic library, a natural language processing system, a voice processing system, an image processing system, an intelligent reasoning system, etc.

[0002]

2. Description of the Related Art In many knowledge processing systems, knowledge is expressed in an expression form called a semantic network (or concept network or associative network). A semantic network is a directed graph that represents the association between words (or concepts and concepts) (Iwanami Information Science Dictionary, edited by Makoto Nagao, Iwanami Shoten, 1990). The nodes in the graph represent words or concepts, and the links between the nodes represent association relationships between words or concepts. Figure 1
5 shows an example of a semantic network. In this network, the nodes are, for example, "car", "truck".
Represents a word such as. Also, the links between the nodes are, for example,
It means that there is a strong recollection from "car" to "truck".
Some recollections are bidirectional. For example, "car" and "aut
There is a two-way link between "o", indicating that both parties easily recall each other.

Semantic networks are also available for searching on the Internet or in electronic libraries. For example, suppose a user wants to search the Internet for a "home page about cars." If "car"
If you search only with the keyword, only homepages that contain the word "car" will be collected, and you will not be able to find all (or a lot) of the information that the user wants. But with the semantic network,
You can find out more information related to cars.
Specifically, search for words linked from "car" on the semantic network, that is, words with a strong association, and search for "auto", "motor", "vehicle", "truc".
When words such as "k" and "Toyota" are found, the home page can be searched by using all of these words as keywords.

Semantic networks can also be used for disambiguation in natural language processing systems, speech recognition systems, image processing systems and the like. For example, OC
In R character recognition system, when converting image data to character data, whether the read word is "population" or "entrance"
It is necessary to judge whether it is. If "birth" or "statistics" appears in the same sentence, it can be judged as "population", and if "exit" or "passage" appears, it can be judged as "entrance". . Such knowledge that "population" and "birth" are strongly related can be expressed in a semantic network and used for disambiguation.

Semantic networks can also be used for reasoning in intelligent reasoning systems. For example, by using a semantic network, it is possible to infer why "if the wind blows, a trough is profitable". Specifically, if a node connected to the "wind" node in the semantic network is found, for example, the path "wind → dust → blind person → shamisen → ... → Okeya", that is, if a sequence of multiple connected nodes is found. , "If the wind blows, the trough is profitable"
You can infer the reason. Also, "wind" and "Okeya"
By finding all the possible paths between, you can discover all the reasons for "if the wind blows, the trough makes."

A knowledge representation called a Bayesian network has also been proposed. Bayesian networks represent the dependence of random variables in the joint probability distribution in a directed graph (Judea Pearl, Probabilistic Reaso
ning in Intelligent Systems: Networks of Plausible
Inference. Morgan Kaufman Publishers Inc., SanMat
eo, California, 1988.). Bayesian networks can also be considered as semantic networks. However, in a general semantic network, there may be loops, but in Bayesian networks, there should be no loops. That is, the Bayesian network is a limited semantic network.

On the other hand, in the fields of information theory and mathematical statistics,
Several quantities have been proposed, called "information measures." For example, the measure of "probabilistic complexity" was proposed by Rissenen (Jorma Rissanen, Fishe.
r Information and Stochastic Complexity, IEEE Tran
sactions on Information Theory, Vol.42., No. 1, p
p.40-47, 1996). Probabilistic complexity is a measure of the amount of information contained in a given data for a given probabilistic model. A so-called "minimum description length principle" was also proposed by Rissenen, and the model with the smallest stochastic complexity of the data is closest to the probability distribution that gave rise to the data, and statistical estimation should select that model. Insist. Stochastic complexity can also be interpreted as the shortest code length (or description length) for describing data by a probabilistic model. Also, for example, a measure called "extended stochastic complexity" was proposed by Yamanishi (Kenji Yamanishi, A Decisi
on-Theoretic Extension of Stochastic Complexity an
d Its Applications to Learning, IEEE Transactions
onInformation Theory, Vol.44, No.4, pp. 1424-1439,
1998). Extended stochastic complexity is a quantity for a model that is also included in given data, but the model may be a parametric class of any real-valued function as well as a probability distribution, and a loss function Is an extension of stochastic complexity in the sense that can be any distortion function, not just logarithmic loss. Besides this, "Akaike's information scale" (Hirotugu Akaike, A New Lo
ok at the Statistical Model Identification, IEEE T
ransactions on Automatic Control, Vol.AC-19, No.
6, pp.716-723, 1974), "Entropy" (Iwanami Information Science Dictionary, edited by Makoto Nagao, Iwanami Shoten, 1990).

Further, conventionally, a method of learning a Bayesian network by using stochastic complexity, that is, a principle of minimum description length has been devised (for example, Joe Suzuki, Yasutaka Otake, Shigeichi Hirasawa, minimum description length). Regarding the selection method of probabilistic models from the viewpoint of criteria and state division, IPSJ Transactions, Vol.33, No.11, pp.1281-1289, 1992).

Further, in Japanese Patent Laid-Open No. 11-96177, an ontology capable of recognizing various relations of words is dynamically generated, and is required for processing a large number of documents over a wide target area. A kind of semantic network automatic generation technique of generating a term dictionary sufficiently containing information to be disclosed is disclosed. Specifically, for each word obtained by morphological analysis of a document, a primary statistic indicating the importance of its appearance (for example, the number of occurrences of an individual word with respect to the total number of words included in all documents) Ratio) is obtained, and some words having large primary statistics are selected as related words. Next, using each of the selected related words as a node, a graph in which a directed link is drawn from the node of the word representatively representing the target area to each node of the related word is initially generated. Next, for all combinations of two nodes of the nodes of the generated graph, co-occurrence statistics, which are statistics for simultaneous appearance of two words of each combination, are calculated. Here, the co-occurrence statistic is calculated as a ratio of the total number of documents (or paragraphs or sentences) in which the two words simultaneously appear to the total number of documents (or paragraphs or sentences) in which the two words appear. Then, the graph is converted based on the calculated co-occurrence statistic, a relation label is attached to the link, and a term dictionary is generated.

[0010]

Semantic networks are widely available knowledge, but in the past, their construction and creation depended on humans. It has at least two problems. One problem is that the scale of a semantic network is usually quite large, so the cost of creating it is enormous.
Another problem is that human-defined knowledge inevitably contains a lot of arbitrariness.

Further, using stochastic complexity,
A method of learning a Bayesian network has been proposed, but it has been possible to create only a limited meaning network that "there should be no loops".

Further, in the technique disclosed in Japanese Patent Laid-Open No. 11-96177, a graph in which a directed link is established from a node of a word representing a target area to each node of a related word is initially generated. Since the graph is converted based on the calculated co-occurrence statistic and the relation label is given to the link, only a limited meaning network called a tree-structured network having a representative word as a root node can be created. In other words, as with the method of learning Bayesian networks, we cannot construct semantic networks with loops. Further, the co-occurrence statistic showing the ratio of the total number of documents (or paragraphs or sentences) in which the two words A and B simultaneously appear to the total number of documents (or paragraph or sentence) in which the two words A and B appear is , The value is the same for both word A and word B. In Japanese Laid-Open Patent Publication No. 11-96177, since graph conversion is performed based on such co-occurrence statistics, whether or not a directed link from word A to word B is possible and word B to word A is possible.
It is not possible to independently decide whether or not to have a directed link to each of them based on the statistical degree of co-occurrence.

An object of the present invention is to provide a semantic network automatic creation apparatus for statistically processing a plurality of texts to automatically create a semantic network.

Another object of the present invention is to provide a semantic network automatic creation apparatus capable of creating a semantic network having a loop.

Another object of the present invention is to automatically create a directional link from one node to another node and a directional link in the opposite direction thereof independently based on the degree of statistical co-occurrence. Providing a production device.

[0016]

A means for automatically creating a meaning network according to a first aspect of the present invention inputs a plurality of texts into a storage part for storing a meaning network composed of words,
Morphological analysis is performed on the input text, and from the text subjected to morphological analysis, a co-occurrence frequency of words and words, a frequency of occurrence of words, a statistics unit that statistics the total number of texts, and words and words from the statistics unit. Enter the co-occurrence frequency, word occurrence frequency, and total text number, and based on the input word and word co-occurrence frequency, word appearance frequency, and total text number, A calculation unit that calculates the strength of recollection of each word using an information measure, and one word is expressed as one node, and for each expressed node, the word from that node to other nodes is expressed. Referring to the strength of the recollection to the word from the calculation unit, if the strength of the referred recollection is greater than or equal to a predetermined threshold value, a directed link is established from that node to another node, and the directed link is A directed graph that has And a creation unit configured to output to the storage unit as a network, wherein the calculation unit, as the amount of information measure
One text with probabilistic complexity
Is regarded as one data, and any first word
The strength of the second word in the text
Probability of data when focusing on whether or not a word appears
Complexity and text in which the first word appears
If we focused on whether the second word appeared in the group,
Probabilistic Complexity and First Word of Combined Data
Second word appears in a text group in which does not appear
Probabilistic Complexity of Data Focusing on Whether
Calculated as the difference from the sum of tees.

[0017]

An apparatus for automatically creating a semantic network according to a second aspect of the present invention inputs a plurality of texts into a storage unit that stores a semantic network consisting of concepts, performs morphological analysis on the input texts, and performs morphological analysis. The word sense disambiguation is performed on the performed text, and the co-occurrence frequency of concepts and concepts, the appearance frequency of the concept, and the statistical part that statistics the total number of texts from the text that has been subjected to the word sense disambiguation, and the concept from the statistical part And the concept co-occurrence frequency, the concept appearance frequency, and the total number of texts are input. Based on the input concept and concept co-occurrence frequency, the concept appearance frequency, and the total number of texts, the concept A calculation unit that calculates the strength of recollection from one to another concept using an information measure, and one concept is expressed as one node, and for each expressed node, from the concept of that node to another No Referring to the strength of recollection to the concept of from the calculation unit, if the strength of the referred recollection is greater than or equal to a predetermined threshold value, a directed link is established from that node to another node, and the directed link is A calculation unit that outputs the stretched directed graph to the storage unit as a semantic network consisting of concepts , and the calculation unit is a stochastic computer as an information measure.
Use plexity and add one text to one
Data from the first concept to the second concept.
The second concept appears in the text
Probabilistic completion of data when focusing on whether or not
In texts where Kisity and the first concept appear
Data when focusing on whether or not the second concept appears
The stochastic complexity and the first concept of
Whether a second concept appeared in a group of texts
And the sum of the probabilistic complexity of the data
Calculate as the difference of.

[0019]

In addition to the configuration of the first or second aspect of the invention, the semantic network automatic creation apparatus according to the third aspect of the present invention further includes an interface section for receiving a designation of a node in the semantic network from the user, and the interface section to the user. A search unit for inputting a designated node and a semantic network from a storage unit for finding a partial semantic network including a node designated by a user in the input semantic network; and a partial semantic network from the searching unit. And a display unit for displaying the input partial meaning network.

In the apparatus for automatically creating a meaning network of the present invention configured as described above, the co-occurrence frequency of words and words (or concepts and concepts) is statistically calculated from the text of natural language, and the information amount scale is used. Then, the statistical strength of co-occurrence between words (or concepts and concepts) is calculated, and this is used as the strength of recall between words (or concepts and concepts). Build a network automatically.

Since the semantic network is automatically constructed, the number of steps required for its construction can be greatly reduced. In addition, the meaning network is constructed objectively by calculating the strength of recollection between words (concepts and concepts) based on the data on a solid scale of theoretical basis. can do.

Further, since the semantic network is created based on the strength of recall between words using the information amount scale, it is possible to create a semantic network having a loop. That is, it is possible to create a more general semantic network than the conventional learning method of Bayesian network.

Furthermore, the strength of recollection from a certain word to another word (a certain concept to another concept) is defined as the appearance / non-occurrence of another word (concept) in the text group in which the word (concept) appears. Since the statistic for appearance and the statistic for appearance and non-occurrence of other words (concepts) in the text group in which the word (concept) does not appear are taken into consideration, calculation is performed from one node to another node. The directed link and the directed link in the opposite direction can be created independently based on the statistical degree of co-occurrence.

[0025]

BEST MODE FOR CARRYING OUT THE INVENTION A first embodiment of an automatic network creation device of the present invention will be described. FIG. 1 shows its configuration, and FIG. 2 shows its processing flow. The semantic network automatic creation device 10 includes a storage unit 1, a statistics unit 2, a calculation unit 3, and a creation unit 4.

The statistics section 2 inputs a plurality of texts from an input device such as a keyboard or a storage device such as a floppy disk device (not shown) (step S1), and performs morphological analysis on these texts (step S2). Further, the frequency of co-occurrence of words, the frequency of appearance of words, and the total number of texts are statistically calculated from those texts (step S
3). Here, the text means a text written in a natural language such as Japanese or English. One text is
For example, one document, one paragraph in the document, one sentence in the paragraph, and the like. Further, the word-to-word co-occurrence frequency is the number of texts in which a word and a word appear together. The word appearance frequency is the number of texts in which the word appears.

Next, the calculation unit 3 inputs the word-to-word co-occurrence frequency, the word appearance frequency, and the total number of texts from the statistical unit 2,
Based on these data, the strength of statistical co-occurrence between words is calculated using an information amount scale, and the strength of association between words is calculated (step S4). The strength of recall between words is generally asymmetric. Specifically, the strength of recollection from word A to word B is usually different from the strength of recollection from word B to word A.

Next, the creation unit 4 expresses each word as a node, and with respect to each node, the calculation unit 3 refers to the strength of recollection from the word of the node to the word of another node, When the strength of recall is equal to or greater than a predetermined threshold,
A directed link is established from that node to another node (step S5). The directed graph thus created is output to the storage unit 1 as a semantic network (step S6).

Hereinafter, each part will be described in more detail.

The statistics section 2 inputs a plurality of texts written in Japanese, English, or the like. FIG. 3 shows an example of Japanese text to be input. FIG. 4 shows an example of the result of morphological analysis performed on the text of FIG. That is,
In Japanese, text is divided into words by morphological analysis. FIG. 5 shows an example of input English text. FIG. 6 shows an example of the result of morphological analysis performed on the text of FIG. That is, in English, a word is converted into a prototype by morphological analysis.

The statistic section 2 stores the morphologically analyzed text as
Considered as a set of words. The statistical unit 2 statistically processes the morpheme-analyzed text to obtain the word-to-word co-occurrence frequency, the word appearance frequency, and the total number of texts. FIG. 7 shows an example of the processing of the statistical unit 2. FIG. 7A shows an example of the morphologically analyzed text, which is 1 in total from text1 to text10.
There is 0 text. Word1 and the like listed next to the characters text1 to text10 are words included in the text, and there are five types from word1 to word5.
In this case, the statistical unit 2 calculates 10 as the total number of texts. Further, if f (w) represents the appearance frequency of the word w, the appearance frequency as shown in FIG. 7B is obtained for each word. Furthermore, if f (x, y) represents the co-occurrence frequency of the word x and the word y, the co-occurrence frequency as shown in FIG. 7C is obtained.

The calculation unit 3 inputs the word-to-word co-occurrence frequency, the word appearance frequency, and the total number of texts from the statistical unit 2. Next, for each word s, the strength of recall from that word s to another word w is calculated. When calculating, an information measure is used.

As an example, a method of calculating the strength of recall from the word s to the word w when the probabilistic complexity is used as the information amount scale will be described. What is required for calculation are the co-occurrence frequency f (s, w) of the words s and w, the appearance frequency f (s) of the word s, the appearance frequency f (w) of the word w, and the number of texts n.

The calculator 3 regards one text as one data. First, the probabilistic complexity of data in the case of paying attention to whether or not the word w appears in the text is calculated. This is called probabilistic complexity for an independent model of data. Specifically, it is given by the following equation. nH (w) + (1/2) log (n / 2π) + logπ (Equation 1) where H (w) = − P (w) logP (w) −P (¬w) logP (¬w) (Equation 2) P (w) = f (w) / n (Equation 3) P (¬w) = 1-P (w) (Equation 4) However, assuming that the base of the logarithm is 2, 0log0 = 0 (same below). π is 3.1416.

Next, the text is divided into two groups, one in which the word s appears and one in which the word s does not appear. In each of the divided groups, the probabilistic complexity of the data when paying attention to whether or not the word w appears is calculated, and the sum of the two is calculated.
This is called stochastic complexity for the data dependence model. Specifically, it is given by the following equation. [F (s) H (w | s) + (1/2) log {f (s) / 2π} + logπ] + [(nf (s)) H (w | ¬s) + (1/2 ) Log {(nf (s)) / 2π} + logπ] (Equation 5) where H (w | s) =-P (w | s) logP (w | s) -P (¬w | s) log P (¬w | s) (Equation 6) P (w | s) = f (w, s) / f (s) (Equation 7) P (¬w | s) = 1-P (w | S) (Equation 8) H (w | ¬s) =-P (w | ¬s) logP (w | ¬s) -P (¬w | ¬s) logP (¬w | ¬s) ( Formula 9) P (w | ¬s) = f (w, ¬s) / f (¬s) = {f (w) -f (w, s)} / {n-f (s)} (Equation 10) p (¬w | ¬s) = 1-P (w | ¬s) (Equation 11)

Next, the difference between the stochastic complexity for the independent model and the stochastic complexity for the dependent model is calculated. It can be said that the greater the difference in the stochastic complexity, the stronger the degree of statistical co-occurrence from the word s to the word w. In particular, P (w | s)>
It can be said that there is a positive co-occurrence when P (w) and a negative co-occurrence when P (w | s) <P (w). In practice, the difference in stochastic complexity is further divided by the total number of texts, as shown in the following equation. S (s → w) = (1 / n) [(Equation 1)-(Equation 5)] (Equation 12) P (s) = f (s) / n (Equation 13) P (¬s ) = 1-P (s) (Equation 14) When Equation 12 is rearranged, the following equation is obtained. S (s → w) = H (w) -P (s) H (w | s) -P (¬s) H (w | ¬s)-(1 / 2n) log [{f (s) (n -F (s)) π} / 2n] (Equation 15)

Here, it should be noted that the strength of the recall from the word s to the word w depends on the word s from the word w.
It is a point that is usually different from the strength of recollection. In other words, the strength of recall between words is asymmetric. This corresponds to the fact that, for example, the word "Kurosawa" is strongly associated with "Samurai", but conversely, the word "Samurai" is not always associated with "Kurosawa." FIG. 8 shows a calculation example of the recall strength from the word word1 to the word word2 in the example of FIG. 7, and FIG. 9 shows a calculation example of the recall strength from the word word2 to the word word1. In this way, the strength of recollection from the word word1 to the word word2 (0.379) is equal to the strength of recollection from the word word2 to the word word1 (0.3).
60) is larger than that in the example of FIG.
If d1 appears, word2 always appears, whereas
This is based on the phenomenon that if word2 appears, word1 does not necessarily appear.

Next, the creation unit 4 and the storage unit 1 will be described. The creation unit 4 expresses a word as a node. Further, the creating unit 4 establishes a link between the nodes of the words having high recall strength while referring to the strength of recall from word to word from the calculation unit 3. Specifically, for example, if the strength of recall from the word s to the word w is larger than the threshold value, a directed link is established from the node of the word s to the node of the word w. In this way, repeat the operation of linking for all words,
Build a semantic network. The storage unit 1 stores the semantic network created in this way.

FIG. 10 shows a processing example of the creating unit 4. First,
A list of all words is prepared (step S11) and put in the list WL1 (step S12). Then list W
The first word w1 of L1 is extracted (step S14),
Put all word list in list WL2 (step S1
5), the following processing is repeated for the word w1.

First, the first word w2 in the list WL2 is taken out (step S17), and if it is not the same word as the word w1 (NO in step S18), the words w1 to w2 are extracted.
The calculation unit 3 refers to the strength of recollection (step S1
9) If the strength of recall exceeds the threshold (step S2)
If YES at 0), a directional link is established from the word w1 to the word w2 (step S21). If the strength of recollection does not exceed the threshold value (NO in step S20), word w1 to word w
Do not link to 2. When the processing for one word in the list WL2 is completed, the same processing is performed for the next word in the list WL2, and this is repeated for all the words in the list WL2.

When the above-mentioned processing is completed for one word in the list WL1 (YES in step S16), the next word is extracted from the list WL1 (step S14).
The same process as described above is performed for the word. This is repeated for all the words in the list WL1 (step S1).
3).

Next, a second embodiment of the meaning network automatic creation apparatus of the present invention will be described. FIG. 11 shows its configuration, and FIG. 12 shows its processing flow. The meaning network automatic creation device 20 includes a storage unit 1, a statistics unit 2,
It is composed of a calculation unit 3 and a creation unit 4.

The statistics section 1 inputs a plurality of texts from an input device, a storage device or the like (not shown) (step S1),
Morphological analysis is performed on these texts (step S2-1), word sense disambiguation is further performed on those texts (step S2-2), and then the co-occurrence frequency of concepts and concepts from those texts, The appearance frequency of the concept and the total number of texts are statisticized (step S3). Next, the calculation unit 3 inputs the frequency of concept co-occurrence, the frequency of concept appearance, and the total number of texts from the statistics unit 2, and based on these data, the strength of recollection from concept to concept Calculation is performed using the scale (step S4). Next, the creation unit 4 expresses the concept as a node, refers to the strength of the recollection from the concept to the concept from the calculation unit 3, and links the concept to the concept with the strength of the recollection being a predetermined threshold value or more. A semantic network is created by stretching (step S5), and this is output to the storage unit 1 (step S6).

The second embodiment differs from the first embodiment in that the first embodiment creates a semantic network of words, whereas the second embodiment consists of concepts. This is the point of creating a semantic network. The concept here is intuitively the meaning of a word.
Therefore, the statistical unit 2 further performs word sense disambiguation on the text subjected to the morphological analysis. For example, in the text examples of FIGS. 5 and 6, "fly" has the meaning of "fly" and "fly". In the disambiguation of word sense disambiguation, the meaning of "fly" in this sentence is determined to mean "fly," that is, the concept represented by "fly" is "fly," depending on the context. After performing word sense disambiguation on the morphological analysis result of FIG.
The result is as follows. In FIG. 13, for example, "fl
“Y2” represents the concept of “fly”, that is, the meaning of “fly”.
The statistical unit 2 then statistics the concept and the co-occurrence frequency of the concept.

The calculation unit 3 inputs the concepts and the co-occurrence frequency of the concepts, the appearance frequency of the concepts, and the total number of texts from the statistical unit 2. Next, for each concept s, the strength of recall from the concept s to another concept w is calculated. The calculation method is the same as in the first embodiment.

The creating unit 4 expresses the concept as a node. Next, the creation unit 4 refers to the strength of recollection from concept to concept from the calculation unit 3 and forms a link between the nodes of the concept having a high strength of recollection. Specifically, for example, the concept s
If the strength of recollection from to concept w is larger than the threshold value, a directed link is established from the node of concept s to the node of concept w.
In this way, the linking operation is repeated for all concepts to create a semantic network. The storage unit 1 stores the semantic network created in this way.

Next, a third embodiment of the meaning network automatic creation apparatus of the present invention will be described. FIG. 14 shows the configuration. The semantic network automatic creation device 30 includes a storage unit 1, a statistics unit 2, a calculation unit 3, a creation unit 4, an interface unit 5, a search unit 6, and a display unit 7.

The third embodiment differs from the first embodiment in that the interface section 5, the search section 6 and the display section 7 are provided. The other parts are the same as in the first embodiment.

The interface section 5 receives a designation of a node in the semantic network from a user through an input device such as a keyboard (not shown). For example, if the user specifies the node "car", it will receive it. The user also inputs the designation of the search range.

The search unit 6 inputs the node designated by the user from the interface unit 5 and the semantic network from the storage unit 1, and finds a partial semantic network including the designated node in the semantic network. Specifically, the search unit 6 associates with the designated node. The association starts with the specified node,
Go to the node connected to that node, and go to the node connected to it. The association is repeated up to the number of times specified by the user in specifying the search range. However, the node that has been asked once by associating is marked so that further associating is not performed from that node.

FIG. 15 shows an example of the semantic network stored in the storage unit 1. FIG. 15 shows a part of a semantic network actually constructed using the automatic semantic network creating device of the present invention based on data of about 9000 newspaper articles of Reuters. According to the present invention, it can be seen that even with a small amount of data, it is possible to create a semantic network that is quite similar to human intuition. When such a semantic network is associated once with a node called "car", the search unit 6 finds a partial semantic network as shown in FIG. 16, and the display unit 7 displays the found partial network. Show users the meaningful network.

FIG. 17 shows a processing example of the search unit 6. The search unit 6 receives the node designation and the designation of how many stages to associate with the user from the user through the interface unit 5 (step S31). Let the designated node be start and the designated number of stages be k. The search unit 6 uses the recursive function Find_Pa
th is executed (step S32). Here, the recursive function Find_Path is a function that takes start and k as arguments and returns a path list Path_list. The list Path_List of the returned paths is displayed on the display unit 7.
Is output (step S33).

FIG. 18 shows a processing example of the recursive function Find_Path. The arguments are node and k. First, nod
e marks Mark as having been associated, Path_Li
st is an empty list (step S41). Argument k is 0
If not (NO in step S42), the list of nodes linked to the node is put in the Linked list (step S43), the first node first is extracted from the Linked list (step S45), and it has never been associated. If it is a node (step S46)
, NO), first and k-1 are called as arguments, and the returned Partial_Path-
The path of List is added to Path_List (step S47). The same process is repeated for the remaining nodes in the Linked list, and when the processes for all the nodes in the Linked list are finished (YES in step S44), nodes are added to the heads of all paths in Path_List to obtain new Path_List.
t is returned (step S48), and the process ends.

Incidentally, in the third embodiment shown in FIG.
Although the interface unit 5, the search unit 6 and the display unit 7 are added to the first embodiment, the interface unit 5, the search unit 6 and the display unit 7 are added to the second embodiment so that the user An embodiment (fourth embodiment) of searching for and displaying a partial semantic network including a node of a concept specified by can be realized.

FIG. 19 is a block diagram showing an example of a computer to which the present invention is applied. The computer A includes a computer main body B including a central processing unit, a main memory, a hard disk device, a floppy disk device, a CD-ROM unit, a display device C, a keyboard D, and a mouse E.
Composed of and. F is a floppy disk, CD-RO
It is a machine-readable recording medium such as M, in which a semantic network automatic creation program is recorded. The meaning network automatic creation program recorded on the recording medium F is read by the computer main body B, and the operation of the computer main body B is controlled, whereby the first embodiment shown in FIG. In the second embodiment shown in FIG. 11, the storage unit 1, the statistics unit 2,
The calculation unit 3 and the creation unit 4 are realized, and further, the interface unit 5, the search unit 6, and the display unit 7 are realized in the third embodiment and the fourth embodiment shown in FIG.

[0056]

As described above, according to the present invention,
It is possible to construct a semantic network by an efficient method that does not require human hands, and it is possible to construct a very objective knowledge because it is created by using an information scale based on data. . Furthermore, it is possible to create a semantic network with loops, and to create directional links from one node to another node and vice versa in an independent manner based on the degree of statistical recall. It is possible.

[Brief description of drawings]

FIG. 1 is a block diagram of a first embodiment of an automatic meaning network creation device of the present invention.

FIG. 2 is a flowchart showing a processing example of the first embodiment of the meaning network automatic creation device of the present invention.

FIG. 3 is a diagram showing an example of input text.

FIG. 4 is a diagram showing an example of a result of morphological analysis on the text of FIG.

FIG. 5 is a diagram showing an example of input text.

FIG. 6 is a diagram showing an example of a result of morphological analysis on the text of FIG.

FIG. 7 is an explanatory diagram of a processing example of a statistical unit.

[Fig. 8] Fig. 8 is a diagram illustrating a specific calculation example of the strength of recall from the word word1 to the word word2.

FIG. 9 is a diagram illustrating a specific calculation example of the strength of recall from the word word2 to the word word1.

FIG. 10 is a flowchart illustrating a processing example of a creation unit.

FIG. 11 is a block diagram of a second embodiment of the meaning network automatic creation device of the present invention.

FIG. 12 is a flowchart showing a processing example of the second embodiment of the meaning network automatic creation device of the present invention.

FIG. 13 is a diagram showing an example of a result of word sense disambiguation for the text of FIG. 6;

FIG. 14 is a block diagram of a third embodiment of the semantic network automatic creation system of the invention.

FIG. 15 is a diagram showing an example of a part of a constructed semantic network.

FIG. 16 is a diagram showing an example of a partially semantic network that can be searched.

FIG. 17 is a flowchart illustrating a processing example of a search unit.

FIG. 18 is a flowchart showing a processing example of a recursive function executed by a search unit.

FIG. 19 is a configuration diagram showing an example of a computer to which the present invention has been applied.

[Explanation of symbols]

1 ... storage unit 2 ... Statistics department 3 ... Calculation part 4 ... Creation department 5 ... Interface section 6 ... Search unit 7 ... Display 10, 20, 30 ... Semantic network automatic creation device

─────────────────────────────────────────────────── ─── Continuation of the front page (56) Reference JP-A-11-96177 (JP, A) JP-A-5-94482 (JP, A) JP-A-2000-56977 (JP, A) Daichi Mochihashi, Yuji Matsumoto , Meaning of association, IPSJ Research Report 99-NL-134-21, Japan, November 26, 1999, Vol. 99, No. 95, p. 155-p. 162 (58) Fields surveyed (Int.Cl. ⁷ , DB name) G06F 17/21-17/30 G10L 15/18 JISST file (JOIS)

Claims (6)

(57) [Claims] 1. A storage unit that stores a semantic network of words, a plurality of texts are input, morpheme analysis is performed on the input texts, and from the texts subjected to the morpheme analysis,
The statistics section that statistics the word-to-word co-occurrence frequency, the word appearance frequency, and the total number of texts, and the word-to-word co-occurrence frequency, the word appearance frequency, and the total number of texts from the statistics section word co-occurrence frequency of the word, the word frequency of, based on the total number of text, for each word, the calculation section calculated using the amount of information measure the strength of the recall from the word to other words There is an information measure
To use probabilistic complexity and
The strike is regarded as one data, and the first word
The strength of recollection of the second word in the
Of the data when paying attention to whether or not two words have appeared
Probabilistic complexity and the text in which the first word appears
Pay attention to whether the second word appears in the strike group
The stochastic complexity of data and the first
The second word appears in the text group in which no word appears
Probabilistic completion of data when focusing on whether or not
A calculation unit that calculates the difference from the sum of kisses, and one word is expressed as one node, and for each expressed node, the strength of recollection from the word of that node to the words of other nodes If the strength of the referred recollection is greater than or equal to a predetermined threshold value by referring to the calculation unit, a directed link from that node to another node is formed, and a directed graph with a directed link is formed of words. An automatic creation device for a semantic network, comprising: a creation unit for outputting to the storage unit as a semantic network. 2. A storage unit that stores a semantic network consisting of concepts, a plurality of texts are input, morphological analysis is performed on the input texts, and word sense disambiguation is performed on the morphologically analyzed texts. And a co-occurrence frequency of concepts and concepts from the text that has been subjected to word sense disambiguation, a statistical part that statistics the frequency of appearance of concepts, and the total number of texts, and a co-occurrence frequency of concepts and concepts from the statistical part, and an appearance frequency of concepts , Input the total number of texts, and based on the input concept and the frequency of co-occurrence of the concept, the frequency of appearance of the concept, and the total number of texts, for each concept, the strength of recall from that concept to other concepts A calculation unit that calculates using an information content scale,
To use probabilistic complexity and
The strike is regarded as one data, and the first concept
The strength of recollection of the second concept of meaning
Of the data when focusing on whether the concept of 2 has appeared
Probabilistic complexity and the text where the first concept emerges
Paying attention to whether the second concept appears in the strike group
The stochastic complexity of data and the first
The second concept appears in the text group where the concept does not appear
Probabilistic completion of data when focusing on whether or not
A calculation unit that calculates as a difference from the sum of chisities, and one concept is expressed as one node, and for each expressed node, the strength of recollection from that node's concept to that of another node When the strength of the referred recollection is equal to or more than a predetermined threshold value, a directed link is established from that node to another node, and the directed graph with the directed link is constructed from the concept. An automatic creation device for a semantic network, comprising: a creation unit for outputting to the storage unit as a semantic network. 3. An interface unit for receiving a designation of a node in a semantic network from a user, a node designated by the user from the interface unit, and a semantic network from a storage unit,
A search unit for finding a partial semantic network including a node designated by the user in the input semantic network; and a display unit for inputting the partial semantic network from the searching unit and displaying the input partial semantic network. 3. The meaning network automatic creation device according to claim 1 or 2 . 4. A storage unit for storing a semantic network of words, inputting a plurality of texts, performing a morphological analysis on the input texts, and selecting a text from the morphological analysis.
Word-to-word co-occurrence frequency, word appearance frequency, statistical part that statistics the total number of texts, word-to-word co-occurrence frequency, word appearance frequency, total text number from the statistics part, and the entered word a co-occurrence frequency of the word, word frequency, based on the total number of text, for each word, met calculator calculates using the amount of information measure the strength of the recall from the word to other words Information scale
To use probabilistic complexity and
The strike is regarded as one data, and the first word
The strength of recollection of the second word in the
Of the data when paying attention to whether or not two words have appeared
Probabilistic complexity and the text in which the first word appears
Pay attention to whether the second word appears in the strike group
The stochastic complexity of data and the first
The second word appears in the text group in which no word appears
Probabilistic completion of data when focusing on whether or not
A calculation unit that calculates as a difference from the sum of kisses , one word is expressed as one node, and for each expressed node, the strength of recall from the word of that node to the words of other nodes is expressed. With reference to the calculation unit, if the strength of the referred recall is equal to or greater than a predetermined threshold value, a directed link is established from that node to another node, and a directed graph with a directed link is composed of words. A computer-readable recording medium in which a program that functions as a creating unit that outputs to the storage unit as a network is recorded. 5. A computer, a storage unit for storing a semantic network of concepts, inputting a plurality of texts, performing morphological analysis on the input texts, and word sense ambiguity for the texts subjected to the morphological analysis. The co-occurrence frequency of the concept and the concept, the appearance frequency of the concept, and the statistical part that statistics the total number of texts from the text that has been resolved and the word sense disambiguation has been performed. Input the frequency and the total number of texts, and based on the input concept and the frequency of co-occurrence of the concept, the appearance frequency of the concept, and the total number of texts, the strength of recollection from that concept to other concepts Is a calculation unit that calculates
To use probabilistic complexity and
The strike is regarded as one data, and the first concept
The strength of recollection of the second concept of meaning
Of the data when focusing on whether the concept of 2 has appeared
Probabilistic complexity and the text where the first concept emerges
Paying attention to whether the second concept appears in the strike group
The stochastic complexity of data and the first
The second concept appears in the text group where the concept does not appear
Probabilistic completion of data when focusing on whether or not
A calculation unit that calculates as a difference from the sum of chisities , one concept is expressed as one node, and for each expressed node, the strength of recall from the concept of that node to the concept of other nodes is expressed. With reference to the calculation unit, if the strength of the referred recall is equal to or greater than a predetermined threshold value, a directed link is established from that node to another node, and the meaning of the concept is a directed graph with directed links. A computer-readable recording medium in which a program that functions as a creating unit that outputs to the storage unit as a network is recorded. 6. A computer, further an interface unit for receiving a designation of a node in a semantic network from a user, inputting a node designated by the user from the interface unit, and inputting a partial semantic network from a storage unit. A search unit for finding a partial semantic network including a node designated by a user in the specified partial semantic network; a display unit for inputting the partial semantic network from the searching unit and displaying the input partial semantic network; Recorded claim 4 or
5. A computer-readable recording medium according to item 5 .