JP3427679B2 - Computer-readable recording medium recording word search device and word search program - Google Patents

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a word search device for efficiently searching for a word from a set of words based on a key assigned to the word and a medium storing a word search program, and particularly to a medium at an arbitrary position. The present invention relates to a word search device that searches a related word at a high speed by designating a certain character, and a medium that stores a word search program for causing a computer to perform such a search.

[0002]

2. Description of the Related Art The process of searching for a word from some key is a basic process in a text information processing system such as a dictionary search or kana-kanji conversion. For that reason, the processing speed and the storage capacity required to retrieve a word from a key greatly influence the performance of such a processing system. Therefore, such processing can be performed faster
Moreover, by realizing from a smaller storage capacity, a very large practical effect can be obtained.

There are already many systems for retrieving a document by using a word as a key. Even in such a document retrieval system, the function of outputting a word as a result of retrieval is not essential but very effective. There is a case.

In many document retrieval systems, the key word is included in the document or is preselected by the system administrator or the document creator. In this case, a user who uses the search system often does not know what the registered key word is. Therefore, in order to assist the user, it is necessary to be able to search the keyword in some way instead of the document.

As a specific method for searching for a keyword, for example, a method for searching a written word from a reading, a method for searching a word containing a certain character or a character string, and a word satisfying an arbitrary regular expression. The method of searching is considered.

Further, in a document search system using keywords, a tree structure called a try (trie index) is often used as an index in order to quickly obtain a desired document or a pointer to the document from a word. This try can be used to search for words at high speed. When searching for a word from a try,
The number of processing steps required is almost proportional to the length of the input character string. Moreover, since the data compression rate is relatively good, the trie is suitable for storing a large number of index words. In addition, when using a try, there is an advantage that if a character string at the beginning of a word is specified, all words starting from the character string can be obtained by a simple process.

However, a word including a partial character at an arbitrary position other than the beginning of the word cannot be searched out at high speed from the try. Here, in these cases, by providing an index separately from the trie representing the word,
It is performed to search a word including the character from a partial character at an arbitrary position.

Now, let us consider a case where a character is used as a search key to search for a word included in the spelling at an arbitrary position from the character. Since a word is composed of a plurality of characters, one certain word is associated with a plurality of search keys (characters). When the search target is linked to a plurality of keys as in this case, the set of search targets is stored in one data structure, and the search key is associated with the value of the pointer in the data structure. Therefore, the required storage capacity is small.

Therefore, the data structure in which the data of the word set can be expressed with a small storage capacity and the word can be specified by the pointer will be considered below. A general data structure for storing a set of words is a record structure for storing a fixed-length or variable-length character string. The word search method using this record structure will be referred to as a "first conventional example". With this record structure, an arbitrary word can be accessed at high speed in a substantially constant time regardless of the total number of words. Therefore, by storing the word set as such a record structure, it is possible to realize high-speed keyword search in the word search system.

In addition to the first conventional example, it is conceivable that the trie index is regarded as word set data and the identification number of the node in the trie corresponding to the end of the word is used as a pointer to the word. . This is the "second conventional example"
I will call it. In a tree structure like try,
All nodes except the root have a unique parent node.
Therefore, by giving the link information to the parent node to all nodes, when one node is designated, the route from the node to the root can be uniquely determined.

[0011]

However, when there is a trie index having a word as a key and a word in the index is searched from some key, the above conventional method has the following problems. There was a point.

In the first conventional example described above, that is, in the method of preparing record structure data for storing a set of words as a fixed-length or variable-length character string separately from the index, a high-speed key-to-word conversion is performed. Such a search can be realized,
New data is required separately from the word set data already stored in the trie format. Therefore, there is a problem in that the storage capacity required is too large to ignore.

The above-mentioned second conventional example, that is,
In the method in which the index is regarded as the data of the word set and the identification number of the node in the try corresponding to the end of the word is used as the pointer to the word, the data for representing the word set is not required as compared with the first conventional example. Although the storage capacity required for the search from the key to the word itself is smaller than that of the first conventional example, the link information to the parent node, which is originally unnecessary for the trie index, is added. Will be done. Therefore, in terms of the search system as a whole, it cannot be said that the second conventional example is significantly improved in storage capacity as compared with the first conventional example.

The present invention has been made in view of the above points, and an object of the present invention is to provide a word search device capable of searching a word at high speed with a small storage capacity. Another object of the present invention is to provide a medium recording a word search program for causing a computer to search for a word at high speed with a small storage capacity.

[0015]

In order to solve the above problems, the present invention relates to a word search device for searching a word from a word set, in which nodes are associated with each other according to a trie format in which nodes are recorded in depth priority order. a word storage means for a set of words is stored, which is the position of the nodes in the word storage means is input, the node of roots will follow in this order, the input position trie said word storing means
Seeking a route until the word search means for tracing all paths after determined route retrieves all words corresponding to the nodes to reach, and outputs a set of the acquired words, before
Keys corresponding to the words included in the word storage means
Store the position of the nodes that make up a word in association with each other
Key index storage means and the key index case
When any key in the payment means is entered, the key index
Node corresponding to the key input from the
Get the set of positions of
Node position searching means for outputting to the word searching means
And the key index storage means is
All the characters that make up the word included in the word storage means, and
Corresponds with the position of the node representing the character, and
Of the characters that make up the words included in the word storage,
All characters except the first and last characters of the word, and
There is provided a word search device characterized by associating the position of a node representing each character with each other .

According to this word search device, when the position information of the storage destination in the word storage means is input, the word search means sequentially follows the tries of the word storage means from the root, and the node at the input position is searched. Is obtained, all the words corresponding to the nodes that reach all the paths after the obtained path are acquired, and the acquired word set is output. Key index storage means is word storage means
The keys that correspond to the words in
The position of the node to be stored is associated and stored. Node position detection
Any key in the key index storage means
When entered, it is entered from the key index storage means.
The set of node positions corresponding to the generated key
Output the set of node positions to the word search means
It Furthermore, the key index storage means is a word storage
Show all the letters that make up the words in the column and each letter
Corresponding to the position of the node that is doing
Of the letters that make up the word included in
And all characters except the last character and each character
It is associated with the position of the existing node. As a result, a plurality of words including the word corresponding to the input position can be searched at high speed, and the storage capacity required for the word storage means can be small.

[0017]

[0018]

[0019]

BEST MODE FOR CARRYING OUT THE INVENTION Embodiments of the present invention will be described below with reference to the drawings. FIG. 1 is a block diagram showing the principle of the present invention. The word search device according to the present invention has a word storage means 1 and a word search means 2.

The word storage means 1 stores a set of words 1a
A trie format in which nodes are recorded in depth priority order (trie
Index). Here, the node depth is the length of the path from the root of the trie to the node. The node depth priority order is the order of nodes that are followed when the search is advanced in the depth direction as much as possible. When comparing a descendant node of a certain node with a younger brother node, the descendant node always has a higher priority (the address value is smaller). Also, when compared with siblings, the node that is the elder brother has a higher priority.

When the location information of the storage location in the word storage means 1 is input, the word search means 2 traces the tries of the word storage means 1 in order from the root, and obtains a route to the node at the input location. , All the words corresponding to the nodes that reach all paths after the obtained path are acquired, and the acquired word set is output. An algorithm for obtaining a word or a group of words including the node from the position of the node will be described below.

FIG. 2 is a flowchart showing an algorithm for obtaining a word or a group of words including a node from the position of the node. This is a process performed by the word search means 2 that has received the node position information. [S1] Consider the transition from the start node in the try of the word storage unit 1. First, prepare an empty label stack to store the label column in the try,
The start node is set to "node A", and the node corresponding to the given node position is set to "node X". [S2] The first child node of the node A is "B1". Since the node B exists at the position next to the node A, the node B1 which is the first child node is immediately obtained. [S3] The adjacent younger brother node of the node B1 is referred to as a node B2. By tracing the trie from the root, the adjacent younger brother node B2 of any node can be immediately obtained. [S4] It is determined whether the node X is equal to the node B1. If they are equal, the process proceeds to step S8, and if they are not equal, the process proceeds to step S5. [S5] It is determined whether the node X is before the node B2. If it is before, the process proceeds to step S6. If not, the process proceeds to step S6. [S6] When the node X is located before the node B2, it can be seen that the node X follows the route to the node B1. Therefore, the label added to the arc from node A to node B1 is stored in the label stack, node B1 is substituted for node A, and the process returns to step S2. [S7] When the node X is located after the node B2, it is understood that the node X does not follow the route of the node B1. Therefore, the node B2 is substituted for the node B1, and step S
Return to 3. [S8] If the node X is equal to the node B1, the node B
All the routes after 1 are traced, and the label strings added to the arcs that are the routes are obtained. [S9] The label string stored in the label stack and the label string obtained from the path after the node B1 are concatenated, and the process ends. This represents the word to be sought.

By the word searching means 2 performing such processing, a set of words including the node can be obtained from the node position. Moreover, since the nodes in the trie are recorded in depth-first order, and by tracing the trie from the root, its adjacent younger brother arc can be specified, so when the node in the trie is specified,
A route including the node can be specified without using the link information to the parent node. The reason will be described below.

First, as a premise, a certain node in the trie is set as a node A, and a node of the first child of the node A is set as B1.
The adjacent younger brother node of the node B1 is B2, and the node B1
Let C be the node of the firstborn child of. Now, assume that the position of the node X is specified and a route from the node A to the node X is to be obtained. At this time, it is assumed that the node X is known to be traced from the node A.

In the trie according to the present invention, since the nodes are recorded in the depth priority order, the node C is between the nodes B1 and B2. If node X is equal to node C, node X is after node B1,
And it is located before the node B2. Therefore, the node X will be traced from the node B1. From the above, it can be seen that when the node X is between the node B1 and the node B2, the node X can be reached by following the transition path from the node A to the node B1.

Further, it is understood that when the node X is after the node B2, the node X does not follow the transition path from the node A to the node B1. In this case, node X
It can be seen that the node A can be reached from the node A by following a path that transits any one of the nodes of the younger brother of the node B1. By repeating the above consideration, the route that can be reached from node A to node X is known.

As a result, when the position of the node is designated, the word including the node can be obtained by transitioning from the root of the trie. Moreover, by referring to the word in the trie by using the position of the node in the trie as a pointer, it is not necessary to provide data representing the word set other than the trie index, and the information storage capacity can be reduced.

It is also possible to output a set of words in response to the input of a search key (character) for a word. Such a word search device will be described below.

FIG. 3 is a block diagram showing the principle of a word search device that inputs a key to a word. This word search device comprises a word storage means 11, a key index storage means 12, a node position search means 13 and a word search means 14. The word storage means 11 and the word search means 14 are
Since it has the same function as the word storage means 1 and the word search means 2 in FIG. 1, the description thereof is omitted here.

The key index storage means 12 stores the key (character) to the word and the position information of the node to which the same label as the key is added in association with each other. In this way, the key index storage means 12 is constructed based on all the correspondences between the keys and the node positions.
When a plurality of words correspond to one key, the key index storage means 12 corresponds to the set of node positions in the key.

When a key to a word is input, the node position searching means 13 searches the key index storing means 12 for a set of corresponding node positions. And
The obtained set of node positions is input to the word search means 14.

After that, the word searching means 14 searches the word storing means 11 for the word assembly corresponding to each node position, and outputs the word assembly. As a result, it is possible to input a character that constitutes a word as a search key and obtain a set of words including the character. That is, it is possible to find a set of words including a partial character at any position other than the beginning of the word.

The key index storage means 12 is
Among the characters forming the word included in the word storage means 11, all the characters except the first character and the last character of the word are associated with the position of the node representing the character in the word storage unit. It may be that is,
This is because the word starting from the first character or the last character of the word can be searched by using the conventional technique, and therefore it is not particularly necessary to manage it in the key index storage means 12. In this way, by omitting the information on the positions of the nodes corresponding to the first character and the last character of the word, the storage capacity required for the key index storage means 12 can be further reduced.

In the word storage means 11, a pointer to information (for example, a document including the word) corresponding to a word and a word corresponding to the pointer are stored as a pair, and the word search means 14 stores A set of pointers to the word represented by each node and information corresponding to the word may be output from the word storage unit 11. Thus, by inputting an arbitrary key (character), a word including the key can be acquired, and further information (document or the like) corresponding to the acquired word can be acquired.

The above is the basic configuration which is the basis of the present invention. Hereinafter, an embodiment of a word search device that is a more specific version of the above configuration will be described. FIG. 4 is a block diagram showing the first embodiment of the present invention. This is a word search device that inputs a word and searches a set of words related to the word.

The word search device according to this embodiment comprises a related word index section 21, a node position search section 22 and a word search section 23. The related word index unit 21 includes a word storage unit 1 shown in the principle configuration of FIG.
In addition to holding the information of 1 and the key index storage means 12, it also holds the correspondence between a word and a set of related words for that word.

Upon receiving the input of a word, the node position searching unit 22 receives a set of node positions of words related to the word (including the word itself), and the word searching unit 23.
To enter. The word search unit 23 searches the related word index unit 21 based on the input set of node positions, acquires the set of related words, and outputs the set.

Before describing the details of the embodiments of the present invention, the terminology related to trees will be newly defined. A tree gives a hierarchical relationship to a set of elements called nodes. In the following, the hierarchical relationship given in the tree will be expressed by words that represent the parent-child relative relationship.
In the tree, there is one node whose descendants are all the nodes including itself. This is called a start node. For every node other than the start node, there is always one parent node. What indicates the parent-child relationship between nodes is called an arc. Then, a node can have an end state as a special state, and is distinguished from other nodes that are not in the end state. In addition, it is assumed that a node having no descendants other than itself is in an end state.

A trie is a kind of tree structure, and each of the paths connecting from the start node to the node in the end state by an arc through an arbitrary number of nodes is one of the sets. The words correspond. In the present description, in the trie, characters, which are elements that form words, are assigned to arcs. Further, the letters corresponding to the respective arcs derived from one node configure the node differently. In a try,
Since there is always one arc that transitions to one node excluding the start node, a set of such a node and arc is regarded as a unit called an edge clause. Below is an example of a trie that represents a set of words.

FIG. 5 is a diagram showing an example of a set of words.
In this word set, there are six words "solution", "analysis", "resolution", "resolution", "development", and "image". A pointer is associated with each word. The pointer points to the position of the identifier set of the document including the word, but in this embodiment, it is also used to point to the node position set of related words. The "resolution" pointer is "T1", the "analysis" pointer is "T2", the "resolution" pointer is "T3", the "resolution" pointer is "T4", and the "development" is The pointer of "" is "T5", and the pointer of "image" is "T6". Based on this set of words, a trie in which nodes in depth-first order are recorded is generated.

FIG. 6 is a diagram showing an example of a trie in which nodes are recorded in the depth priority order. This is the six words “solution”, “analysis”, “resolution”, “resolution”, “development” shown in FIG.
It is a representation of the "image" by a trial. In the figure, nodes 30 to 38 are represented by circles or double circles. Double circles represent end nodes (nodes where the corresponding word exists). The node 30 which is the root becomes the "start node" at the time of word search. The symbols at the lower right of the node in the end state represent the identifiers that mean the set of words related to the words indicated by the node in the end state. Each node 3 other than the start node in the figure
The addresses (positions) of the respective nodes are displayed near 1 to 38. In addition, each node 31 to 38
The arrows connecting the lines are arcs 41 to 48. The character above each arc 41 to 48 (here, one Kanji character) is a label.

When the trie nodes in FIG. 6 are arranged in the depth-first order, the node 31 that has transited the arc 41 having the label of “solution” derived from the root node 30 and the label of “analysis” derived from the node 31. "Image" derived from the node 32 and the node 31 that have transited the arc 42 with
Node 33 that has transited the arc 43 with the label
The order is ...

In this way, the side nodes are recorded in the depth priority order,
At the same time, identify a younger brother's position immediately next to a certain verse. FIG. 7 is a diagram showing an example of a trie index. In this figure, the address where the information on the side clause is stored, the position (address) of the younger brother node immediately below that side clause, the label corresponding to that side clause, the state of the node, and the link information to the corresponding document set are shown. Shows. In this example, the label is one Kanji character. The node status is set to either "end" or "continue" or both. "End" indicates that the word corresponding to the node exists, and "Continue" indicates that the node has children.

By constructing a trie in this way, when a certain "Haraku A" is specified, the eldest son of "Harushi A" and the younger brother of "Hiraku A" immediately next to him are immediately specified. it can.

The operation of this embodiment will be described on the assumption of concrete case data. For the sake of simplicity, the word set shown in FIG. 5 is given, and it is assumed that “resolution” and “resolution” are related to each other.

First, a trie representing all the words is created, and a link to a set of words related to the word is provided at the final node corresponding to the end of the word. This is called try 1. If the word set of FIG. 5 is given, the trie of FIG. 6 is constructed.

Next, a table for associating words with the link information to the set of words is prepared. This is called a related word correspondence table. FIG. 8 is a diagram showing a related word correspondence table. In this table, link information from a word to a set of related words and a set of node positions are associated with each other. In this example, since the "resolution" and the "resolution" are related to each other, the node positions "14, 19" correspond to the link information "T3" and "T4".

Next, the operation of executing the search according to this embodiment will be described with an example. Consider a case where a word is input as a search key and a word set related to the input word is obtained. Based on the assumed case data,
Consider the case where "resolution" is entered as a search key.

First, the node position search unit 22 sets a set of node positions corresponding to the word "resolution" of the search key,
That is, the link information to the related word set can be obtained.
Specifically, the procedure is as follows. The usual way of searching for words in a try is to search for "T" in the try index.
Link information "3" is requested. Then, “14, 19” is obtained as a node position set corresponding to “T3” from the related word correspondence table.

Next, the word search unit 23 obtains a corresponding word set from the node position set obtained by the node position search unit 22. Below, based on the trial of FIG.
A procedure for searching for a corresponding word using one node position as a key will be described.

FIG. 9 is a flow chart showing the word search procedure in the first embodiment. [S11] It is assumed that the node address of the search key is "A". First, a label recording unit for recording a character string of a word is prepared. At the beginning, pay attention to the side clause at the beginning of the try. The buns you are paying attention to is the "current buns"
, And the arc and node included in the side clause are referred to as “current arc” and “current node”, respectively. [S12] It is determined whether the current side clause is equal to "A". If they are equal, the process proceeds to step S13, and if they are not equal, the process proceeds to step S14. [S13] When the address of the current edge node is equal to "A", the label of the current arc is pushed (stored) in the label recording unit, the contents of the label storage unit are output, and the process is terminated normally. [S14] It is determined whether the current node has a younger brother. If it has a younger brother, it proceeds to step S15, and if it does not have it, it proceeds to step S17. [S15] When the current side node has a younger brother, it is determined whether the address of the side node of the adjacent younger brother is "A" or less. If "A" or less, the process proceeds to step S16, and "A"
If not, proceed to step S17. [S16] If the address of the adjacent younger brother's side node is "A" or less, pay attention to the adjacent younger brother's side node and proceed to step S12. [S17] It is determined whether or not the current edge node has children. If the child has a child, the process proceeds to step S18. If the child does not have a child, the processing ends as an illegal end. [S18] If the current side node has children, it is determined whether the address of the longest side node is "A" or less. If "A" or less, the process proceeds to step S19, and "A"
If it is not below, the processing ends as an unauthorized end. [S19] The label of the current arc is pushed to the label recording section, paying attention to the longest node, and the process proceeds to step S12.

Through the above processing, the corresponding related word is output using the node position as a key. As an example, when the address “14” is obtained as the link information to the word,
According to the above algorithm, the corresponding related word is obtained from the tries recorded as shown in FIG.

First, from step S11, "1" is added to "A".
Substitute 4 ”and pay attention to the side node of address“ 0 ”. Leave the label recording area empty. Hereinafter, the edge clause will be identified by its address. That is, the address "0"
The side clause of is called the side clause “0”. Nodes and arcs will be identified in the same way as in the edge clause.

In step S12, the current side clause "0" is not equal to "A (= 14)", so the process proceeds to step S14. In step S14, since the younger brother's side clause exists in the current side clause "0", the process proceeds to step S15. In step S15, adjacent younger brother = 24, A
= 14, and the adjacent younger brother ≤A does not hold, the process proceeds to step S17.

At step S17, since a child side node exists in the current side node "0", the process proceeds to step S18. In step S18, the eldest child = 5 and A = 14, and since the eldest child ≦ A holds, the process proceeds to step S19.

In step S19, the label "solution" of the current arc "0" is pushed to the label recording section, and attention is paid to the longest node "5". Then, the process proceeds to step S12. In step S12 again, the current side clause “5”
Is not equal to “A (= 14)”, step S14
Proceed to.

In step S14, since the younger brother's side clause exists for the current side clause "5", the process proceeds to step S15. In step S15, the adjacent younger brother (= 10) ≦ A
Since (= 14) holds, attention is paid to the adjacent younger brother “10”, and the process proceeds to step S12.

In step S12, the current edge clause "1"
Since "0" is not equal to "A (= 14)", step S
Proceed to 14. In step S14, the current side clause "1
Since there is no brother's side clause for "0", step S1
Proceed to 7.

In step S17, the current side clause "1
Since "0" has a child side clause, the process proceeds to step S18. In step S18, the eldest child = 14 and A = 14, and since the eldest child ≦ A holds, the process proceeds to step S19.

In step S19, the label "image" of the current arc "10" is pushed to the label recording section,
Attention is paid to the eldest son, “14”. And step S1
Go to 2.

In step S12, the current edge clause "1"
Since "4" is equal to "A (= 14)", the process proceeds to step S13. In step S13, the current arc "14"
The label "degree" is pushed to the label recording unit, the contents "solution, image, degree" of the label recording unit are output, and the processing is normally completed.

By the above processing, the node position "1"
From "4", the corresponding word "resolution" can be obtained. Similarly, from the node position "19",
"Resolution" is obtained. That is, the output of “resolution and resolution” is obtained with respect to the input of “resolution”.

In this way, it is possible to perform the related word search processing at high speed with a small storage capacity by using the trie. Next, a second embodiment will be described. Second
The embodiment is a word search device that searches a character for a keyword including the character.

FIG. 10 is a diagram showing a schematic configuration of the second embodiment. The word search device according to this embodiment is
It is composed of a word storage unit 51, a character index unit 52, a node position search unit 53, and a word search unit 54. Of these components, the word storage unit 51, the node position search unit 53, and the word search unit 54 have the functions of the word storage unit 11, the node position search unit 14, and the word search unit 12 shown in FIG. There is. Further, the character index section 52 uses the key index of the key index storage means 13 of FIG. 3 as a specific character index.

First, a set of index words is represented by a trie to form the word storage unit 51. The last node representing the index word is linked so as to follow the set of documents related to the index word. Similar to the first embodiment, the word index (trie index) by the trie of FIG. 6 is configured corresponding to the index word given as shown in FIG. This word index is stored in the word storage unit 51.

Further, a character index that can derive the index word from the characters that form the index word is formed. This character index is the character index part 52.
Stored in. In the case of the index word set of Fig. 5, the set of all the characters that make up the index word is {solution, current, analysis, image, degree,
Power}. For each character in this character set, the position of each character is obtained by following the trie in FIG. 6 from the beginning. A character index is created from the result.

FIG. 11 is a diagram showing an example of a character index. As shown in the figure, a character is associated with an index word containing the character. In this example, the trie 55 is adopted as the data structure of the character index.

It can be seen from FIG. 5 that there are a plurality of index words including the character "solution", that is, "solution", "analysis", "resolution", and "resolution", but the link information to these is 1 representing the character "solution". Only one node position "0" is required. As described above, by using the position of the node of the character in the trie index, the capacity of the character index can be reduced as compared with the case where the node position at the end of the word is used as a pointer to the word.

Now, the operation of executing the search will be described with an example. Consider a case where one character is input as a search key and an index word including this character and a link to a document set are obtained. As a specific example, assume that “image” is input as a search key using the indexes of FIGS. 11 and 6.

First, the node position searching unit 53 finds a link from the character indexing unit 52 to an index word containing a character (“image” in the example) of the search key.
It can be seen from FIG. 11 that the link information “10”, “28”, and “33” can be obtained.

Next, the word search unit 54 obtains the index word and the document set pointed to by the index word from the link information by using the word index. The procedure will be described below.

12 to 14 are flowcharts showing the processing procedure for obtaining a document set in the second embodiment. FIG. 12 shows the processing of steps S21 to S27, FIG. 13 shows the processing of steps S31 to S36, and FIG. 14 shows the processing of steps S41 to S48. The processing contents of each step will be described below. [S21] It is assumed that the link information to the index word to be obtained is given by the side node address of the try of “A”. The label recording part for recording the character string of the index word, the label stack for temporarily saving the character string recorded in the label recording part, and the segment when the label is saved are saved. Prepare each bunsetsu stack for each and clear the contents. Go to the start node and note the edge clause at the beginning of the try. We call the bunsetsu of interest the present bunsetsu. [S22] The current side node is given the link information "A".
And whether or not If they are equal, step S23
If not equal, the process proceeds to step S31 (shown in FIG. 13). [S23] The current node is the given link information "A"
If so, the label of the current arc is pushed to the label recording unit. [S24] It is determined whether the current node is in the end state.
If it is the end state, the process proceeds to step S25, and if it is not the end state, the process proceeds to step S26. [S25] When the current node is in the end state, the contents of the label recording section and the link to the document set corresponding to the current end state node are output. [S26] It is determined whether or not the current node has children. If it has a child, the process proceeds to step S27. If it does not have a child, the process ends normally. [S27] If the current side node has a child, pay attention to the longest node and proceed to step S41 (shown in FIG. 14). [S31] If it is determined in step S22 that the current side clause has a younger brother, the process proceeds to step S32.
If no brother is present, the process proceeds to step S34. [S32] It is determined whether or not the address of the side node of the adjacent younger brother is "A" or less. If it is "A" or less, the process proceeds to step S33, and if it is not "A" or less, the process proceeds to step S34. [S33] If the address of the adjacent younger brother's side node is "A" or less, pay attention to the adjacent younger brother's side node and proceed to step S22 (shown in FIG. 12). [S34] If the address of the adjacent younger brother's node is not equal to or less than "A", it is determined whether the current node has a child. If the child has a child, the process proceeds to step S35. If the child does not have a child, the processing ends as an illegal end. [S35] It is determined whether or not the address of the child's long node is "A" or less. If "A" or less, the process proceeds to step S36.
If it is not "A" or less, the processing ends as an illegal end. [S36] The label of the current arc is pushed to the label recording unit and attention is paid to the long node. And step S
22 (shown in FIG. 12). [S41] It is determined whether or not the current node has a younger brother. If it has a younger brother, the process proceeds to step S42, and if it does not have a younger brother, the process proceeds to step S43. [S42] If the current side node has a younger brother, the contents of the label recording unit are pushed onto the label stack, and the side nodes of the adjacent younger brother are pushed onto the side node stack. [S43] It is determined whether the current node is in the end state.
If it is the end state, the process proceeds to step S44, and if it is not the end state, the process proceeds to step S45. [S44] When the current node is in the end state, the contents of the label recording section and the label of the current arc are connected, and the link to the document set corresponding to the current end state node is output. [S45] It is determined whether or not the current node has children. If it has a child, the process proceeds to step S46, and if it does not have a child, the process proceeds to step S47. [S46] If the current side node has a child, the label of the current arc is pushed to the label recording unit to pay attention to the long side node. Then, the process proceeds to step S41. [S47] It is determined whether the label stack and the side clause stack are empty. If it is not empty, the process proceeds to step S48. If it is empty, the process normally ends. [S48] When the label stack and the side segment stack are not empty, the side stack popped from the label stack to the label recording unit and the side segment popped from the side segment stack is focused on. Then, the process proceeds to step S41.

By the above processing, the character string of the index word and the link from the index word to the corresponding document set are output. As an example, when “10” is obtained as the link information to the index word, the trie of FIG. 7 is followed according to the above algorithm.

First, from step S21, "1" is added to "A".
Substitute “0” and pay attention to the side clause “0”. Label recording section,
Leave both the label stack and the side clause stack empty. In step S22, the current node "0" is not equal to "A (= 10)", so the process proceeds to step S31.

In step S31, since the younger brother exists in the current node "0", the process proceeds to step S32.
In step S32, adjacent younger brother = 24, A = 10
Since the adjacent younger brother ≦ A does not hold, the process proceeds to step 34.

At step S34, since the current node "0" has a child, the process proceeds to step S35.
In step S35, the eldest child = 5 and A = 10,
Since the firstborn ≤ A holds, the label "solution" of the current arc "0" is pushed to the label recording unit, and the first child's side clause "5".
Pay attention to. Then, the process proceeds to step S22.

In step S22 again, the current side clause "5" is not equal to "A (= 10)", so that the process proceeds to step S31. In step S31, the current side clause “5”
In contrast, since the younger brother's side clause “10” exists, step S
Proceed to 32. In step S32, the adjacent younger brother (= 1
Since 0) ≦ A (= 10) holds, the adjacent younger brother “1
Paying attention to "0", the process proceeds to step S22.

At step S22, the current side clause "1
Since "0" is equal to "A (= 10)", the label "image" of the current arc "10" is pushed to the label recording unit. The content of the label recording section at this point is "resolution". It proceeds to step S24. In step S24, the current node "10" is not in the end state, so step S2
Go to 6. In step S26, the current side clause “1
Since "0" has a child, pay attention to the longest child's edge node "14" and proceed to step S41.

At step S41, since the current node "14" has a younger brother, the process proceeds to step S42. In step S42, "resolution", which is the content of the label recording unit, is pushed onto the label stack, and the adjacent younger brother's side node "19" is pushed onto the side node stack. Then, the process proceeds to step 43.

At step S43, the current node "14" is in the end state, so the routine proceeds to step S44. In step S44, "resolution", which is a combination of "resolution", which is the content of the label storage unit, and "degree", which is the label of arc "14", and link information to the document set corresponding to node "14". "T3" is output. Then, the process proceeds to step S45.

In step S45, the current node "14" has no child, so the flow advances to step S47.
In step S47, the label stack and the side clause stack are not empty, so the process proceeds to step S48. In step S48, the "resolution" is popped from the label stack and is taken out into the label recording unit, and the side node "19" is popped out from the side node stack, and the side node "19" is noted. At present, the label stack and the edge stack are empty. Then, the process proceeds to step S41.

In step S41 again, since the current node "19" has no younger brother, the process proceeds to step S43. In step S43, the current node "19" is in the end state, so the process proceeds to step S44. In step S44, "resolution", which is a combination of "resolution", which is the content of the label storage unit, and "force", which is the label of the arc "19".
And outputs "T4" which is the link information to the document set corresponding to the node "19". Then, the process proceeds to step S45.

In step S45, the current node "19" has no children, so the flow advances to step S47.
In step S47, since the label stack and the side clause stack are empty, the processing ends normally.

By the above processing, link information from "10", which is a link to the index word including the character "image", to two index words ("resolution" and "resolution"), and the corresponding document set. (“T3” and “T4”) can be obtained.

Next, a third embodiment will be described.
FIG. 15 is a block diagram showing a schematic configuration of the third embodiment. This is a word search device that searches a keyword that satisfies a regular expression from an arbitrary regular expression.

The word search device according to this embodiment comprises a word storage unit 61, a character index unit 62, a node position search unit 63, a word search unit 64, and a regular expression analysis unit 65. Of these, the word storage unit 61, the character index unit 62, the node position search unit 63, and the word search unit 64 are the word storage unit 51, the character index unit 52, the node position search unit 53, and the word search shown in FIG. It has substantially the same function as the part 54.

When the regular expression search key is input, the regular expression analysis unit 65 analyzes the search key and obtains a word set that matches the search key. At that time, if necessary, the character string is input to the node position searching unit 63, and the node position of each character string is obtained as the return value. Also, the node position is input to the word search unit 64 to obtain a word set.

The operation when the regular expression analysis unit 65 executes a search will be described below. In the trie, when two nodes “N1” and “N2” are designated, “N1” can be traced from the node “N1” and cannot be traced from the node “N2”. The subtree of the trie that is defined as is referred to as a subtrie defined by "N1" and "N2".

Here, the following three procedural functions "F
1 ”,“ F2 ”, and“ F3 ”are defined. The processing of these functions is performed by the regular expression analysis unit 65. First, the function “F1” will be described. The function “F1” is a function that takes the subtrie “T” and the regular expression “R” as arguments and returns the set “S” of character strings as a value.

16 and 17 are flowcharts showing the processing procedure of the function "F1". FIG. 16 shows the step S
51 to S58, and FIG. 17 shows step S61.
7 shows the processing of S68. [S51] It is determined whether the beginning of the regular expression "R" starts with the fixed character string "B" or the character set "B". If it starts with a fixed character string or the like, the process proceeds to step S52, and if it does not start with the fixed character string, the process proceeds to step S61. [S52] When the beginning of the regular expression "R" starts from the fixed character string "B" or the character set "B", the regular expression excluding the portion "R" from "R" is "R1", and Process. [S53] It is determined whether or not the subtrie “T” can be traced by inputting the character string or the character set “B”. If it can be traced, the process proceeds to step S55, and if it cannot be traced, the process proceeds to step S54. [S54] The error is returned as the value "S" of the function "F1", and the process ends. [S55] If "T" can be traced from "B", it is determined whether the regular expression "R1" is empty. If it is empty, the process proceeds to step S56, and if it is not empty, step S58.
Go to. [S56] If the regular expression “R1” is empty, it is determined whether or not the destination following the input “B” on the subtrie “T” is the node in the end state. If it is the end state, the process proceeds to step S57, and if it is not the end state, the process proceeds to step S54. [S57] If the node following the subtrie "T" input "B" is in the end state, the character string "B" is changed to the function "F".
The value is returned as the value "S" of "1", and the process ends. [S58] If the regular expression “R1” is not empty, and the subtrie “T” is traced with the character string “B” as input,
The remaining subtries that can be further traced are designated as "T1". Subtrie "T1" as an argument to the function "F1"
And the regular expression "R1" are passed, and the character string "B" and the function "F
Concatenate each of the character string sets that are the evaluated values of "1",
The obtained character string set is returned as the value "S". Then, the process ends. [S61] It is determined whether or not the fixed character "C1, C2, C3, ..." Is included in the regular expression "R". If the confirmed character is included, the process proceeds to step S62, and if not, the process proceeds to step S68. [S62] Determined character Ci (where i = 1,
2, 3, ...) In the trie, the number of occurrences of nodes is Ai, and the character “Cj” having the minimum value among A1, A2, A3 ,. Determined from the output of. [S63] For each of the node positions “Pi (where i = 1, 2, 3, ...)” representing the character “Cj”, the regular expression “R” includes the character “Cj” at the end. A part of "R" is defined as a regular expression "R1" and the remaining regular expressions are defined as "R2", a finite state automaton "M" that accepts "R1" is created, and the process proceeds to step S64. When the processing is completed for all the node positions “Pi”, the process proceeds to step S67. [S64] Finite state automaton "M", subtrie "Ti", node position "P" as an argument to the function "F2"
i ”is passed, and the character string“ s ”that is the value obtained by evaluating the function“ F2 ”is obtained. [S65] It is determined whether the character string "s", which is the value of the function "F2", is normally output. If the output is normal, step S6
6, the output is not normal (when the character string "s" is an error), the process proceeds to step S63. [S66] The subtrie that can be traced from the node position "Pi" is set to "Ti". The subtrie "Ti" and the regular expression "R2" are passed as arguments to the function "F1", and the character string "s" and each element of the character string set "S1" which is the value evaluated from the function "F1" are concatenated. Then, a set of character strings is calculated and pushed as a character string set "S". Then, the process proceeds to step S63. [S67] The character string set "S" is output, and the process ends. [S68] The number of characters "N" that satisfies "R" is obtained, the sub-try "T" and the number of characters "N" are passed to the function "F3" as arguments, and the character string set that is the value obtained by evaluating the function "F3" is obtained. , And output as a character string set “S”, and the process ends.

Next, the function "F2" will be described. The function "F2" takes the finite state automaton "M", the subtrie "T", and the node position "P" as arguments, and the character string "s".
Is a function that returns as a value.

FIG. 18 is a flow chart showing the processing procedure of the function F2. [S71] When the node position "P" is designated in the subtrie "T", the route to the node is uniquely determined. Therefore, the label string "L" obtained by tracing the subtrie "T" to the node position "P" is used as the input of the finite state automaton "M". [S72] It is determined whether the label string "L" is accepted by the finite state automaton "M". If it is accepted, the process proceeds to step S73, and if it is not accepted, the process proceeds to step S74. [S73] When the label string "L" is accepted by the finite state automaton "M", the label string "L" is output and the process ends. [S74] If the label string "L" is not accepted by the finite state automaton "M", an error is output and the process ends.

Next, the function "F3" will be described. The function "F3" takes the subtrie "T" and the number of characters "N" as arguments, and returns the set "S" of character strings as a value. FIG. 19 shows
It is a flow chart which shows a processing procedure of function F3. [S81] When the start node is in the end state and the number of characters "N" can be "0", an empty character string is pushed to the character string set "S". [S82] The subtrie "T" is traced by the number of characters "N", and the set of label strings "L" whose node is the end state is obtained, and "L" is pushed to the set of character strings "S". To do. [S83] It is determined whether "S" is empty. If "S" is empty, the process proceeds to step S85, and if "S" is not empty, the process proceeds to step S84. [S84] If "S" is not empty, "S" is output and the process ends. [S85] If "S" is empty, an error is output and the process ends.

When the function is defined as above, the function "F
By passing a trie of a word index and an arbitrary regular expression as an argument to “1” and evaluating the function “F1”, a set of words conforming to the given regular expression can be obtained.

As a specific example, let us consider a case of searching for a word that matches the regular expression “? Ide? A?”. In the regular expression, "?" Means a wildcard corresponding to zero or more arbitrary characters. The intent of the search is to search for words that include the character string "idea", taking into consideration fluctuations in notation such as "idea" with respect to "idea." As described above, by being able to search for a word that matches the regular expression, it is possible to reduce omission of search.

For example, consider a case where the following word index is stored in the word storage unit 61. Figure 20
FIG. 16 is a diagram showing an example of a trie 66 according to the third embodiment. In this example, one katakana character is given as the label of each side clause.

FIG. 21 is a diagram showing an example of a character index in the third embodiment. In this character index, the "number of appearances" and the "corresponding node position" are associated with each character.

Here, the evaluation procedure of the value of the function “F1” when the word index trie “T” and the regular expression “? Ide? A?” (R) are passed as arguments to the function “F1”. explain.

In step S51, the regular expression "R"
Starts from the fixed character string, the process proceeds to step S61. In step S61, the regular expression "R" includes the confirmed character ("a", "i", "de"), so the process proceeds to step S62.

In step S62, the three fixed characters "R" {C1, C2, C3} = {A, Y,
The number of appearances of the node representing D} is obtained. It can be seen from the character indexes in FIG. 21 that they are “5”, “2”, and “2”, respectively.

In step S63, "i" is taken as the character having the minimum value among these. Then, all node positions {P1, P
The following processing is performed for 2} = {10, 120}.

In step S63, the regular expression "?
Oh? , "R1" for "? Eye" and "R2" for "de?
Oh? , And make a finite automaton that accepts “R1”, “M”.

FIG. 22 is a transition diagram of the finite state automaton 67 when "a" is taken up. As shown in the figure, when “a” and “a” appear consecutively in the middle of the character string (may be the first), the node transits to the end state node.

In step S64, the finite state automaton "M", the subtrie "T", and the node position "P1" are passed as arguments to the function "F2", and the function "F2" is evaluated. Now, in step S71 of the processing of the function "F2" (shown in FIG. 18), the label string "L" obtained until the subtrie "T1" reaches the node position "P1 (= 10)" is "eye". .

In the judgment at step S72, it is understood that "L" is accepted by the finite state automaton "M". In step S73, the label string “L” (= “eye”)
Is output and the processing of the function “F2” is completed.

The value of the function "F2" is the character string "eye". Therefore, in step S64, this is set to "s". In step S65, the character string "s" (= "eye")
Is not an error, the process proceeds to step S66.

In step S66, the function "F1" is added to the subtrie "T1" as an argument and the regular expression "R2" (= "data").
? A? ") And evaluate the function" F1 ". Hereinafter, the function "F1" will be referred to as a function "F1 '" or the like.

FIG. 23 is a diagram showing a subtrie "T1" that can be traced from the node position P1 (= 10). This subtrie 68 is an extraction of all the routes corresponding to the descendants of the node position P1 (= 10) in the trie 66 of FIG.

Now, in step S51 of the function "F1 '", the head of the regular expression "R'" begins with the fixed character "B '" (= "de"). Step S5
In step 2, "R1 '" is set to "?
Proceed to 53.

In step S53 of the function "F1 '", the subtrie "T'" is input with the character "B '" (= "de") as an input.
Can be traced, the process proceeds to step S55. In the determination of the function “F1 ′” in step S55, the regular expressions “R1 ′” and “?
Proceed to 58.

In step S58 of the function "F1 '", the sub-trie "T'" is input with the character "B '" (= "de") as an input.
The remaining subtries that can be further traced are referred to as "T1 '". Subtrie "T1 '" and regular expression "R1'" (= "? A?") As an argument to the function "F1"
To evaluate the function "F1". Hereinafter, this F1 will be referred to as F1 ″.

Now, step S51 of the function "F1""
, The regular expression “R ″” (= “? A?”) Does not start from the fixed character string, so step S61.
Go to.

In step S51 of the function "F1", the fixed character "C1""(=" A ") is included in the regular expression" R "" (= "? A?"). Therefore, the process proceeds to step S62.

In step S62 of the function "F1 ''",
“C1 ″” (= “A”) corresponds to “Cj ″”, and a set of node positions {P1 ″, P2 ″} = {3
0,80} is obtained.

With respect to “C1 ″” (= “A”), the processing of step S63 and subsequent steps of the function “F1 ″” is performed. In step S63 of the function "F1", "?" Is taken as "R1" and "?" Is taken as "R2". "R
Create a finite state automaton "M""that accepts 1"".

In step S64 of the function "F1", the finite state automaton "M"", the subtrie" T "", the node position "P" are added to the function "F2" as arguments.
1 ″ ”(= 30) is passed and the function“ F2 ”is evaluated.
Hereinafter, this function “F2” will be referred to as a function “F2 ″” or the like.

Now, step S71 of the function "F2 ''"
, The sub-trie "T""at the node position" P "
The label string “L ″” obtained up to 1 ″ ”(= 30) is input to the finite state automaton“ M ″ ”as“ A ”.

In step S72, "L ''" (=
It can be seen that "a") is accepted by the finite state automaton "M ''". Step S73 of the function "F2 ''"
Then, the label string “L ″” (= “A”) is output, and the processing of the function “F2 ″” is completed.

The value "A" of the function "F2 ''" is a character string. In step S64 of the function "F1", this is set to "S". Step S65 of the function "F1""
Since the character string “S ″” (= “A”) is not an error, the process proceeds to step S66.

In step S66 of the function "F1", the subtrie of tracing from the node position "P1" (= 30) is set to "T1"", and the subtrie" T1 '"is added to the function" F1 "as an argument. '"And the regular expression" R2 "" (=
Pass "?" And evaluate the function "F1". Hereinafter, this function “F1” will be referred to as a function “F1 ′ ″” or the like.

Now, step S5 of the function "F1 '""
1, the beginning of the regular expression “R ′ ″” (= “?”) Does not start from the fixed character string, so step S61
Go to.

At step S61 of the function "F1 '",
Since no fixed character is included in the regular expression “R ″ ′” (= “?”), The process proceeds to step S68. Function "F
In step S68 of "1"'","R'"" (=
The number of characters “N ′ ″” that satisfies “?”) Is unlimited.
The character string set which is a value obtained by passing the sub-try “T ′ ″” and the number of characters “N ′ ″] as arguments to the function“ F3 ”is {“ ”,“ rhythm ”}. , As the value of the function “F1 ′ ″”.

A character string set {"", which is a value obtained by evaluating the character string "s""(=" A ") and the function" F1 '"".
A set of character strings which connect each element of "rhythm"} is {"a", "arism"}. In step S66 of the function "F1", this is pushed to the character string set "S"". The contents of "S" at the moment are {"A",
"Arism"}. Then, the process proceeds to step S63.

The processing for the node position "P1" (= 30) is completed by the processing of the function "F1""after step S63. Next, similarly, the process for “P2 ″” is performed, the process returns to step S63 again, and as a result, “S ″” = {“a”, “arism”, “ia”} is output.

Returning to step S58 of the function "F1 '",
The value of the function "F1 '" by connecting the character "B'" (= "de") and each element of the value "", "arism", "ia"} of the function "F1"" Returns {“Dare”, “Dearism”, “Dear”}.

Returning to step S66 of the function "F1", the character string "s" (= "eye") and each element of the value {"Dare", "Dearism", "Dear"} of the function "F1 '" , And push {“idea”, “idea rhythm”, “idea”} to the character string set “S”. The content of "S" at the present moment is {"idea", "idea rhythm", "idea"}. It proceeds to step S63.

In step S63 of the function "F1",
Node position "P2" (= 12)
The following processing is performed for 0). In step S64 of the function "F1", the finite state automaton "M", the subtrie "T", the node position "P" are added to the function "F2" as arguments.
2 ”is passed, and the character string“ s ”(=“ neoeye ”), which is the value obtained by evaluating the function“ F2 ”, is obtained.

In step S65 of the function "F1", there is no error in the character string "s" (= "neoeye"), so the flow advances to step S66. In step S66 of F1, "P2"
The subtrie that can be traced from (= 120) is "T
2 ”, and the function“ F1 ”has a subtrie“ T ”as an argument.
2 ”and the regular expression“ R2 ”(=“ de? A? ”) Are passed, and the function“ F1 ”is evaluated. A character string set {"Dare", "Dearism"} is obtained from the function "F1", {"Neo idea", "Neo idea rhythm"} is pushed to "S", and the process proceeds to step S63. The contents of "S" at the moment are {"idea", "idea rhythm", "idea", "neo idea", "neo idea rhythm"}
Is.

At step S63 of the function "F1", the processing is completed for all the node positions "P1" and "P2", so at step S67 the character string set "S" (= {"idea", "idea rhythm"). , "Idea", "Neo-idea", "Neo-ideaism"}) function "F
The value is returned as a value of "1", and the process ends.

As described above, from the word index try and the regular expression “? Ide? A?”, The word set {“idea”, “idea rhythm”, “idea”, “neo idea”, “neo idea rhythm” } Is obtained.

The following modifications can be considered for the above-described principle configuration or embodiment. FIG. 24 shows the fourth
3 is a block diagram showing a schematic configuration of the embodiment of FIG. This is provided with a plurality of related word index parts 21 in the first embodiment (shown in FIG. 4).

In this embodiment, a related word index is stored in each of the two related word index sections 71 and 72. When a word is input, the node position searching unit 73, the related word indexing units 71 of both sides,
Obtain the position set of nodes from 72. The set of positions of the nodes is determined by which of the related word index parts 71 and 7
The word search unit 74 together with the information indicating whether it is acquired from
Passed to.

The word searching unit 74 is the node position searching unit 73.
Based on the set of node positions received from, the set of related words is acquired from the related word index units 71 and 72 and output.

FIG. 25 is a block diagram showing a schematic structure of the fifth embodiment. This is a concrete embodiment of the fourth embodiment (shown in FIG. 24). In this embodiment, a reading index portion 71a and a notation index portion 72a are provided. Reading index part 71a
Stores the set of notation words represented by the notation in a trie format in which nodes are recorded in depth-first order, and also includes the notation words and the reading words that compose the character string corresponding to each notation word. Stored in association with a set of node positions. The notation index unit 72a stores a set of words represented by readings in a trie format in which nodes are recorded in depth-first order, and constructs reading words and character strings of notation words corresponding to the reading words. It is stored in association with the set of positions of the nodes that are operating.

The node position searching unit 73a and the word searching unit 74
The a has the same function as the node position searching unit 73 and the word searching unit 74 of the fourth embodiment. FIG. 26 shows the sixth
3 is a block diagram showing a schematic configuration of the embodiment of FIG. This corresponds to the related word index parts 71 and 7 of the fourth embodiment.
It is an increase of 2.

In this embodiment, a large number of related word index parts 81a, 81b, 81c, ... Are provided. When the node position search unit 82 receives a word input, each of the related word index units 81a, 81b, 81
From c, ..., Receive a set of positions of the corresponding nodes. The word search unit 83 acquires and outputs a set of related words from the related word index units 81a, 81b, 81c, ... Based on the set of node positions received from the node position search unit 82.

As described above, according to the present invention, a set of words is stored according to a trie format in which nodes are recorded in the depth-first order to form a word storage unit and a word in a trie. By using the node position as a value that can uniquely identify the word in the try, it is possible to specify a route including an arbitrary node without using the link information to the parent node. Further, by referring to the words in the trie by using the position of the node in the trie as a pointer, data other than the trie index that separately represents the word set becomes unnecessary.
As a result, the required storage capacity is significantly smaller than in the case of the prior art.

The above-mentioned processing functions can be realized by a computer. In that case, the processing contents of the functions that the system construction supporting device should have are described in a program recorded on a computer-readable recording medium, and by executing this program on a computer, the above-described processing is realized on the computer. To be done.
Computer-readable recording media include magnetic recording devices and semiconductor memories. When distributing in the market, CD-ROM (Compact Disc Read Only Memor)
y) or a floppy diskette or other portable recording medium for storing and distributing the program, or storing it in a storage device of a computer connected via a network,
It can also be transferred to other computers via a network. When the program is executed by the computer, the program is stored in a hard disk device or the like in the computer, loaded into the main memory and executed.

[0139]

EXAMPLE As a working example of the present invention,
The case of the fifth embodiment and the case of the conventional technique will be quantitatively compared.

FIG. 27 is a diagram showing a correspondence relationship between a written word and a set of reading words corresponding thereto. "Words expressed by notation" are shown on the left side of the figure, and "corresponding readings" are shown on the right side. For example, when the notation is “A”, it may be called “Arufa” or “er”.

FIG. 28 is a diagram showing a correspondence relationship between a reading word and a set of written words corresponding to the reading word. In the figure, "words represented by readings" are shown on the left side, and "corresponding notations" are shown on the right side. For example, when the reading is "A", the notation is "A""A""A""A""A"
There are many such as "Yes".

27 and 28 show the first eight words, the total number of written words is 93,452 words.
The reading word consists of 68,819 words. FIG. 29 is a diagram showing the information amount of the index part in the fifth embodiment. Thus, using the present invention, as a result of creating an index from a notation word to a reading word set and an index from the reading word to the notation word from the correspondence data of the notation word and the reading word, the trie is 1,257,57.
9.0 bytes, the pointer table is 494,085.
0 bytes, the whole index is 1,751,664.0
It has become a storage capacity of bytes.

Therefore, the first conventional example in the description of the conventional technique, that is, a method of preparing record structure data for storing a set of words as a fixed-length or variable-length character string separately from the index, Stored information for performing functions.

FIG. 30 is a diagram showing the amount of information in the first conventional example. In the first conventional example, the text and its reference table are 2,042,570.5 bytes, the trie is 1,257,579.0 bytes, the pointer table is 494,085.0 bytes, and the entire index is 3,7.
It has a storage capacity of 94,234.5 bytes.

A second conventional example in the description of the prior art, that is, the trie index is regarded as word set data, and the identification number of the node in the trie corresponding to the end of the word is used as a pointer to the word. By the way, we stored the information to perform the same function.

FIG. 31 is a diagram showing the amount of information in the second conventional example. In the second conventional example, the trie index contains about 2,469,996.5 bytes and a pointer
The table can be expected to require 494,085.0 bytes and the total index storage capacity of 2,964,081.5 bytes. In the prediction, it is assumed that the data width required for linking to the parent node is 2.5 bytes from the approximate size of the trie data, and the number of try nodes is N, and the data capacity L required for linking to the parent node is L.
Was calculated as L = N × 2.5 (bytes).

Based on the above results, the fifth embodiment and the conventional example were compared. FIG. 32 is a diagram showing a result of comparison of information amount between the fifth embodiment and the conventional technique.

From this comparison result, in the fifth embodiment of the present invention, the index excluding the tries is 19.5%, the overall index is 46.2%, and the second conventional example is compared with the first conventional example. It can be seen that the try requires only 50.9% and the overall index requires only 59.1% of the storage capacity. Therefore, it can be said that the present invention achieves a significant reduction effect of the required storage capacity.

[0149]

As described above, according to the present invention, a set of words is stored according to a try format in which nodes are recorded in depth-first order, and the positions of the nodes forming the words in the try are checked during the try. Since the word is used as a value that can be uniquely identified, a route including an arbitrary node can be specified without using the link information to the parent node. Moreover, since the position of the node in the trie is used as a pointer to refer to the word in the trie, it is not necessary to prepare data representing the word set separately from the word set in the trie format. As a result, the required storage capacity is significantly smaller than in the case of the prior art. In particular,
-Words included in the word storage means
Of the letters that make up the first and last letters of a word
Of all the characters except for and the node that represents each character
Since it corresponds to the position, the key index
The storage capacity required for the storage means can be reduced.

[0150]

[Brief description of drawings]

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

FIG. 2 is a flowchart showing an algorithm for obtaining a word or a set of words including a node from the position of the node.

FIG. 3 is a principle configuration diagram of a word search device that inputs a key to a word.

FIG. 4 is a block diagram showing a first embodiment of the present invention.

FIG. 5 is a diagram showing an example of a set of words.

FIG. 6 is a diagram showing an example of a trie in which nodes are recorded in the depth priority order.

FIG. 7 is a diagram showing an example of a trie index.

FIG. 8 is a diagram showing a related word correspondence table.

FIG. 9 is a flowchart showing a word search procedure in the first embodiment.

FIG. 10 is a diagram showing a schematic configuration of a second embodiment.

FIG. 11 is a diagram showing an example of a character index.

FIG. 12 is a flowchart (No. 1) showing a processing procedure for obtaining a document set in the second embodiment.

FIG. 13 is a flowchart (No. 2) showing a processing procedure for obtaining a document set in the second embodiment.

FIG. 14 is a flowchart (No. 3) showing a processing procedure for obtaining a document set in the second embodiment.

FIG. 15 is a block diagram showing a schematic configuration of a third embodiment.

FIG. 16 is a flowchart (part 1) showing the processing procedure of the function F1.

FIG. 17 is a flowchart (part 2) showing the processing procedure of the function F1.

FIG. 18 is a flowchart showing a processing procedure of a function F2.

FIG. 19 is a flowchart showing a processing procedure of a function F3.

FIG. 20 is a diagram showing an example of a trie according to the third embodiment.

FIG. 21 is a diagram showing an example of a character index according to the third embodiment.

FIG. 22 is a transition diagram of a finite state automaton when “a” is selected.

FIG. 23 is a diagram showing a sub-trie “T1” that can be traced from the node position P1 (= 10).

FIG. 24 is a block diagram showing a schematic configuration of a fourth embodiment.

FIG. 25 is a block diagram showing a schematic configuration of a fifth embodiment.

FIG. 26 is a block diagram showing a schematic configuration of a sixth embodiment.

FIG. 27 is a diagram showing a correspondence relationship between a written word and a set of reading words corresponding to the written word.

FIG. 28 is a diagram showing a correspondence relationship between a reading word and a set of written words corresponding to the reading word.

FIG. 29 is a diagram showing the information amount of the index part in the fifth embodiment.

FIG. 30 is a diagram showing the amount of information in the first conventional example.

FIG. 31 is a diagram showing the amount of information in the second conventional example.

FIG. 32 is a diagram showing a result of comparison of information amount between the fifth embodiment and the conventional technique.

[Explanation of symbols]

1 word storage means 2 Word search means 11 word storage means 12-key index storage means 13 node position search means 14 word search means

─────────────────────────────────────────────────── ─── Continuation of the front page (56) References JP-A 63-12043 (JP, A) JP-A 1-214970 (JP, A) Masuichi Hiroshi, 3 others, Full-text search system using morphological analysis And its application, IPSJ Research Report Natural Language Processing NL-102 -3, July 21, 1994, Vol. 94, No. 63, p. 17-24 (58) Fields surveyed (Int.Cl. ⁷ , DB name) G06F 17/30

Claims (6)

(57) [Claims] 1. A word search device for searching for a word from a word set, wherein a word storage unit stores a set of words associated with a node according to a trie format in which nodes are recorded in depth priority order, and When the position of the node in the word storage means is input, the tries of the word storage means are traced in order from the root, the route to the node at the input position is obtained, and all the routes after the obtained route are traced. All the words corresponding to the nodes reached by, and a word search means for outputting a set of the acquired words, a key corresponding to the words included in the word storage means,
Corresponds and stores the positions of the nodes that make up the word
Key index storage means and any key in the key index storage means is input.
Then, it is input from the key index storage means.
And get the set of node positions corresponding to the key
Output the set of node positions to the word search means
And a node position search means for performing the key index storage means in the word storage means.
Show all the letters that make up the included words and each letter
Corresponding to the position of the existing node, and the word storage means
Of the letters that make up the word included in
And all characters except the last character and each character
A word search device characterized in that it is associated with the position of a node that is present . 2. When a regular expression is input, the input positive
The regular expression is analyzed, and the node position search means
If you pass a character in the expression and receive a set of node positions
Together, the word search means
You pass it a set of words and it matches the regular expression you have entered.
It also has a regular expression analysis means that outputs a set of matching words.
The word search device according to claim 1, wherein: 3. The word storage means pairs a word with each word.
It is stored as a pair with the pointer to the corresponding information.
The word search means is provided for each node from the word storage means.
A collection of pointers to the words represented by and the information corresponding to the words
The word search according to claim 1, wherein the word is output.
apparatus. 4. The word storing means has a depth
According Tra i format de is recorded, corresponding to the node
It stores a set of attached words, and
Corresponds to the set of node positions of related words related to words
When a word is input, it is input from the word storage means.
A set of node positions of words related to the selected word,
The set of acquired node positions to the word search means
It further has a node position searching means for outputting
The word search device according to claim 1. 5. The word storage means is represented by notation.
Nodes are recorded in the order of depth,
Notation words and each notation, as well as being stored in a try format
Nodes of reading words that make up a character string corresponding to a word
A reading index that stores it in association with the set of positions
The storage means and the set of words represented by the reading
Stored in a trie format in which nodes are recorded in order of priority
Together with the reading word and the notation word sentence corresponding to each reading word
Correlate with the set of node positions that make up the string
It consists of the notation index storage means that is stored and
5. The word search device according to claim 4, wherein
Place 6. A computer is searched for a word from a word set.
A computer that records a word search program for
In a readable recording medium, according to a trie format in which nodes are recorded in depth priority order.
A single word that stores the set of words associated with the node.
The word storage means and the position of the node in the word storage means are input.
Then, following the trial of the word storage means in order from the root
The route to the node at the input position,
To a node that follows all routes after the route
Acquires all the corresponding words and outputs the acquired word set.
Forces word search means, a key corresponding to the words contained in the word storage means, each
Corresponds and stores the positions of the nodes that make up a word
All the words that make up the words contained in the word storage means
Corresponds characters to the position of the node that represents each character
And sentences forming the words included in the word storage means
All letters except the first and last letters of the word
Corresponds all characters to the position of the node that represents each character
Key index storage means to be attached , and any key in the key index storage means is input.
Then, it is input from the key index storage means.
And get the set of node positions corresponding to the key
Output the set of node positions to the word search means
A word search program for making a computer function as a node position search means
A computer-readable recording medium on which a ram is recorded.