JP2013088923A - Important query extraction device, important query extraction method and important query extraction program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide an important query extraction device capable of determining and outputting a query with a high degree of importance, in association with input queries.SOLUTION: An important query extraction device 10 comprises: a query acquisition part 11 for acquiring queries; a query log storage part 17 for storing query logs; a first node setting part 121 for setting the acquired queries to a first node; a second node setting part 122 for extracting a query including a word included in the queries of the first node from the query logs, and setting the extracted query to a second node; a score setting part 13 for adding a directed edge between the nodes, and setting its score; and a query output part 14 for identifying an important node on the basis of the score, and outputting the query as an important query. The score setting part 13 sets a score of the directed edge going from one node to the other node, on the basis of the number of words included in the other node and a numerical value showing scarcity of words common to the nodes.

Description

  The present invention relates to an important query extraction device, an important query extraction method, and an important query extraction program for extracting an important query from a plurality of queries.

A search engine having a keyword search function is used as a system for searching information on the Internet.
At this time, the user who performs the search may not be able to input an appropriate keyword. For this reason, a method for presenting a more appropriate query to a user who has input a keyword has been proposed (for example, see Patent Document 1).

  In Patent Document 1, a query rank is calculated based on a query occurrence frequency and a user satisfaction score related to the query, and a query having a higher rank is extracted. A query having a high similarity to the input query and having a high query rank is provided as a modified query.

JP 2008-535090 A

  However, in Patent Document 1, even if a query includes a spelling error, if the input frequency of the query is high, the query is determined to have a high rank. Therefore, the query desired by the user is not necessarily determined to have a high rank. There was a problem.

  An object of the present invention is to provide an important query extraction device, an important query extraction method, and an important query extraction program that can determine and output a query having high importance in relation to an input query.

  The important query extraction device of the present invention includes a query acquisition unit that acquires a query including a word that is a search keyword, a query log storage unit that sequentially stores the query acquired by the query acquisition unit as a query log, and the query acquisition A query including a word included in the query of the first node is extracted from a query log stored in the query log storage unit and a first node setting unit that sets the query acquired in the first node Set by the second node setting unit for setting the query to the second node, a score setting unit for adding a directed edge between the nodes, and setting a score for each directed edge, and the score setting unit. A query output unit that identifies an important node based on the score and outputs a query set in the node as an important query, and the score setting unit includes: Score directed edges directed from square node to the other, and the number of words contained in the other node, to said setting based on the value representing the scarcity of words common to each node.

According to the present invention, the query log storage unit sequentially stores a keyword search query transmitted from a terminal device via a network such as the Internet and stores it as a query log. For example, when a query is input from each terminal device to a search server that searches a web page using a keyword, the query is transferred from the search server to the important query extraction device of the present invention and stored in the query log storage unit. The Therefore, the important query extraction device of the present invention may be incorporated in various search servers such as a web page, or may be configured to be able to perform data communication with the search server.
Then, when a query is input from the terminal device, the first node setting unit sets the input query to the first node. The second node setting unit extracts a query including a word included in the query of the first node from the query log storage unit and sets the query as a second node.
For example, when two words, a proper noun “A” that is a singer name and a word related to the attribute of the singer “song”, are input as input queries, the first node setting unit displays “A song”. A query including these two words is set as the first node. The second node setting unit extracts a query including the word “A” and a query including the word “song” and sets the extracted query as a second node. For example, the query set as the second node includes various queries that have been input including the proper noun “A” in the past, such as “A”, “A, image”, “A, lyrics”, and “lyrics, Various queries that have been input including “music” in the past, such as “music”, are extracted, and a plurality of second nodes are set by these.

The score setting unit adds a directed edge between the nodes and sets the score of the directed edge. At this time, when there are a plurality of second nodes, the directed edge is set not only between the first node and each second node but also between the second nodes. At this time, the score setting unit sets the score of the directed edge between the two nodes according to the direction of the directed edge. That is, two directed edges between the first node and the second node are set, a first directed edge from the second node toward the first node and a second directed edge in the opposite direction. A score is set for each directed edge.
The score of each directed edge between two nodes is high when the rarity of words common to the two nodes is high. The numerical value representing the rarity is, for example, the number of queries including the word in the query log stored in the query log storage unit, and it can be determined that the rarity is higher as the number of queries is smaller. For example, the word “A” that is the name of a singer has a smaller number of query histories than an ordinary noun such as “song”. This is because the word “song” is often input together with other singer names. The rarity may be created, for example, by creating a database of words and setting a numerical value representing the rarity for each word.

  The score of the directed edge is set in consideration of the number of words in the node in the direction toward the directed edge. For example, the score is set to be lower as the number of words is larger, and the score is higher as the number is smaller. That is, when two words “A (singer name), song” are input as the first node, a node having only the singer name “A” as the second node, and “A, B, C” Assuming a node in which three singer names are set, the other singer names B and C are unnecessary as search keywords for the user who has input the singer name A and the song as a query, and it causes noise. . For this reason, a query with only the singer name “A” is more important. Since “B, C” are both proper nouns, if it is attempted to set a score based only on the rarity of a word, it is difficult for the two queries to produce a difference in score. On the other hand, if the number of words is added to the score calculation, the smaller the number of words, the higher the score, and the higher the score of the query for the singer name “A” alone.

  As described above, when a word having a high rarity is input, it can be estimated that the user most demands information related to the word. Even if words with a high degree of rarity are included, if the number of words in the query is large, other words may become noise and an appropriate search result may not be obtained. Accordingly, a query including the word with a small number of words can be determined as the most important query, and according to the present invention, an important query can be accurately extracted from the query log stored in the query log storage unit. The important query extracted in this manner is output to the user terminal device or the like by the query output unit. For this reason, the user who operates the terminal device can obtain an important query such as a correct notation of a proper expression, which does not include a spelling error or a misrepresentation.

  A user who operates the terminal device can select an important query as a query candidate. Accordingly, when there is no appropriate search result of the input query, an appropriate search result can be obtained by selecting the important query output from the important query extraction device of the present invention and searching again. it can.

In the important query extraction device of the present invention, when there is a node composed only of the most rare word among the words included in the input query, the score setting unit is directed to the node. It is preferable that the score of the directional edge is set to be the highest, and the query output unit outputs a query of a node having the highest score of the directional edge as an important query.
The score setting unit increases the edge score as the query has a smaller number of words, and increases the edge score as the word is more rare. For this reason, among the words included in the input query, the score of the edge toward the node of the query including only one of the rarest words is the highest value. Therefore, the query output unit outputs a query of a node having a high score as an important query.
For this reason, if a query of only proper nouns exists in the history of the query log storage unit like the singer name A, the query is output as an important query, and if the user searches using the important query, Search results can be obtained.

In the important query extraction apparatus of the present invention, the score setting unit includes H_ {i, j} as the score of the edge from the node i to the node j, N_q_j as the number of words constituting the query set in the node j, and the query log. A numerical value indicating the number of queries including the word w in the storage unit is DF (w), and a value obtained by adding the reciprocal of DF (w) regarding the words included in the nodes i and j is Σ_ {w∈q_i∩q_j} 1 / DF In the case of (w), the edge score is preferably obtained by the following equation (1).
[Equation 1]
H_ {i, j} = 1 / N_q_j * (Σ_ {w∈q_i∩q_j} 1 / DF (w)) …… (1)

If this formula (1) is used, since the number of words of node j is the denominator, the score becomes smaller as the number of words increases. For words common to each node, the reciprocal of DF (w) indicating the number of queries including the word is added for the common words. Accordingly, the score becomes larger as the number of queries is smaller, that is, the word is more rare.
Therefore, the score increases when the number of words is small and the rareness of words common to each node is high, and it can be determined that the importance of the query increases as the score increases.

  The important query extraction method of the present invention is an important query extraction method for extracting an important query by a computer, and the computer includes a query log storage unit in which a history of the query is stored, and a word as a search keyword is included. From the query acquisition step of acquiring a query, the first node setting step of setting the query acquired in the query acquisition step as a first node, and the history of queries stored in the query log storage unit, A query including a word included in the query is extracted, a second node setting step for setting the extracted query as a second node, a directional edge is added between the nodes, and a score for each directional edge is set. And an important node is identified based on the score set in the score setting step and the score set in the score setting step. A query output step of outputting the query set to as an important query, wherein the score setting step calculates the score of the directed edge from one node to the other node, and the number of words included in the other node. And setting based on a numerical value representing the rarity of words common to each node.

  The important query extraction program of the present invention includes a query acquisition step for acquiring a query including a word as a search keyword in a computer, and a first node setting step for setting the query acquired in the query acquisition step as a first node; A query including a word included in the query of the first node is extracted from a query history stored in a query log storage unit in which a query history is stored, and the extracted query is set as a second node. 2-node setting step, adding a directional edge between the nodes, setting a score of each directional edge, and identifying an important node based on the score set in the score setting step, An important query extraction program that executes a query output step that outputs the query set for that node as an important query The score setting step includes a score of a directed edge from one node to the other node, the number of words included in the other node, and a numerical value representing the rarity of words common to each node; It is characterized by setting based on

  Also in these important query extraction methods and important query extraction programs, the same effects as the important query extraction device can be obtained.

The block diagram which shows the structure of the search system concerning embodiment of this invention. The figure which shows the directed graph of this embodiment. The flowchart which shows the important query extraction process of this embodiment.

Hereinafter, embodiments of the present invention will be described with reference to the drawings.
FIG. 1 shows a search system 1 using an important query extraction device 10 of the present invention.
The search system 1 includes an important query extraction device 10, an information search device 20 that collects and searches web page information, and a terminal device 30. The important query extraction device 10 and the information search device 20 and the terminal device 30 are configured to be communicable via the Internet (not shown). The important query extraction device 10 and the terminal device 30 constitute a search assistance system.

The Internet is the Internet based on a general-purpose protocol such as TCP / IP. Each terminal device 30 is preferably connectable to the Internet while moving. For this reason, the terminal device 30 is connected to the Internet via a mobile phone network or connected to the Internet via a wireless LAN (LOCAL AREA NETWORK).
Note that the configuration for connecting the terminal device 30 to the important query extraction device 10 and the information search device 20 is not limited to the Internet, and a communication line network in which a plurality of base stations capable of transmitting and receiving information using a wireless medium constitute a network. Any configuration that transmits and receives data can be used, such as a network such as a broadcast network, or a wireless medium itself that directly receives data.

  The important query extraction device 10 and the information search device 20 have a hardware configuration including a storage unit (not shown) such as a RAM (RANDOM ACCESS MEMORY) and a ROM (READ ONLY MEMORY), a control unit such as a CPU (CENTRAL PROCESSING UNIT), It is comprised with the server apparatus provided with. The RAM has a temporary area where data and the like can be temporarily stored, and the ROM stores an OS (OPERATING SYSTEM: basic software), programs for controlling various servers, various applications, various data, and the like. The CPU reads programs stored in these storage means and performs various processes according to the programs. In particular, the important query extraction apparatus 10 reads out an important query extraction program stored in the storage unit and executes an important query extraction method described later.

[Information retrieval device]
As shown in FIG. 1, the information search device 20 includes a crawler unit 21, a web page database (WEB page DB) 22, and a search engine 23.
The crawler unit 21 is a search robot that circulates web pages on the Internet, collects the contents thereof, and registers them in the WEB page DB 22.
The search engine 23 performs full-text search or the like. When a search keyword is input from the terminal device 30, the search engine 23 searches the keyword using the WEB page DB 22 and outputs the search result to the terminal device 30. .
Such an information retrieval apparatus 20 is conventionally known and will not be described in detail.

[Important query extractor]
As illustrated in FIG. 1, the important query extraction device 10 includes a query acquisition unit 11, a node setting unit 12, a score setting unit 13, a query output unit 14, a query log storage unit 17, and a query relation storage unit 18. The node setting unit 12 includes a first node setting unit 121 and a second node setting unit 122.

The query acquisition unit 11 acquires a query input from the terminal device 30 to the information search device 20 from the information search device 20 and stores the query in the query log storage unit 17. For this reason, the history of past queries input from various terminal devices 30 to the information search device 20 is stored in the query log storage unit 17.
Further, the query acquisition unit 11 outputs the query input from the information search device 20 to the node setting unit 12 this time.

The node setting unit 12 sets a directed graph node indicating the relationship between the queries acquired by the query acquisition unit 11 and stores the node in the query relationship storage unit 18.
The processing in the node setting unit 12 will be described based on a specific example.
First, an example of a directed graph (query relationship model) set in the node setting unit 12 and stored in the query relationship storage unit 18 will be described with reference to FIG.
The directed graph shown in FIG. 2 is an example when a query including two words of a singer name ABCDEF and a song is input.

That is, the first node setting unit 121 of the node setting unit 12 sets the input query “ABCDEF, song” in the first node 41.
Next, the second node setting unit 122 extracts a query including the words “ABCDEF” and “song” included in the first node 41 from the query log storage unit 17 and sets the query as the second node. In the example of FIG. 2, as the second node, a node 42 having only the singer name “ABCDEF”, a node 43 having two words “ABCDEF, image”, a node 44 having two words “ABCDEF, lyrics”, “ A node 45 composed of two words “lyrics and music” and a node 46 composed only of “music” are set.
A query 47 “ABGDEF” in which the singer name is misspelled is also stored in the history. When a similar query is also extracted as in the prior art, this query 47 is also set as the second node and an edge to be described later is provided, but in this embodiment, no edge is set in the similar query. For this reason, in FIG. 2, “x” marks are given to the edge portions.

Directed edges 51 to 70 are set between the nodes 41 to 45. That is, between the two nodes, two edges, a directed edge from one node to the other node and a directed edge in the opposite direction, are set.
The node setting unit 12 creates an effective graph as shown in FIG. 2 and stores it in the query relation storage unit 18.

The score setting unit 13 sets the score of each directed edge 51 to 70. The score setting unit 13 sets the score of the directed edge from one node to the other node based on the number of words included in the other node and the rarity of words common to each node.
Specifically, the score setting unit 13 sets the highest score of an edge toward a directed node composed of only words having the highest rarity among words included in the input query.
Word rarity is determined by the number of queries that contain the word in the query history. That is, it can be determined that the scarcity is higher as the number of queries is smaller. Comparing proper nouns such as singer names with common nouns such as songs, images, and lyrics, proper nouns usually have fewer queries that contain the word and are more rare. For this reason, in the graph of FIG. 2, the directed edges 51, 57, and 59 that go to the node 42 that is composed of only a rare word (singer name ABCDEF) have the highest score.

An example of a specific score calculation formula in the score setting unit 13 is as shown in the following formula (2).
[Equation 2]
H_ {i, j} = 1 / N_q_j * (Σ_ {w∈q_i∩q_j} 1 / DF (w)) …… (2)

  Here, H_ {i, j} represents the score of the directed edge from node i to node j. N_q_j represents the number of words constituting the query q_j set for the node j. Furthermore, DF (w) is a numerical value indicating the rarity of the word w, and in this embodiment, is the number of queries including the word w in the query history of the query log storage unit 17. Σ_ {w∈q_i∩q_j} 1 / DF (w) is obtained by calculating the reciprocal of DF (w) related to the word w common to the query q_i set to the node i and the query q_j set to the node j. Indicates the value added for the common words.

An example of calculating the score at the directed edges 51 and 52 will be described.
The directed edge 51 is an edge from the node 41 (the node i in the formula 2) toward the node 42 (the node j in the formula 2). Then, the number of words (N_q_j) of the query set in the node 42 is “1” because only the singer name ABCDEF is present. Further, when the number of queries including the singer name ABCDEF is 1000 in the history of the query log storage unit 17, DF (ABCDEF) = 1000.
Since the only word common to the nodes 41 and 42 is ABCDEF, Σ_ {wεq_i∩q_j} 1 / DF (w) = 1/1000. Then, the score H_ {i, j} is 1/1 * 1/1000 = 0.001.

On the other hand, in the case of the directed edge 52, the number of words of the query set in the node 41 is “2”.
Since the only word common to the nodes 41 and 42 is ABCDEF, Σ_ {wεq_i∩q_j} 1 / DF (w) = 1/1000. Then, the score H_ {i, j} is 1/2 * 1/1000 = 0.0005, which is smaller than the directed edge 51.

The directed edges 57 and 59 have the same score as the directed edge 51 because the common word is the same in ABCDEF, and the number of words in the node in the direction toward the edge is “1”.
Further, the directed edges 53 to 62 have the same score as the directed edge 52 because the common word is the same in ABCDEF and the number of words in the node in the direction toward the edge is “2”.

On the other hand, the words common to the two nodes in the directed edges 63 to 70 are “music” and “lyrics”. Such common nouns also increase the number of queries stored in the query log. If the number of queries for the word “song” is 10,000, DF (song) = 10000. Therefore, the score of the directed edges 67 and 69 is 1/1 * 1/10000 = 0.0001, and the score of the directed edges 63, 64, 68, and 70 is 1/2 * 1/10000 = 0.00005. .
If the number of queries for the word “lyric” is 8000, DF (lyric) = 8000, and the score of the directed edges 65, 66 is 1/2 * 1/8000 = 0.000500625.
The score setting unit 13 stores the score of each directed edge in the directed graph of the query relationship storage unit 18. Specifically, the score is set by regarding the score H_ {i, j} as a matrix.

The query output unit 14 searches the directed edge having the highest value among the directed edge scores set by the score setting unit 13, and determines that the node in the direction to which the directed edge is important is important. In the example of FIG. 2, the score of the directed edges 51, 57, 59 is the highest value. Therefore, the query output unit 14 determines that the node 42 is important. Therefore, the query output unit 14 outputs the query “ABCDEF” of the node 42 to the terminal device 30 as an important query.
For this reason, the user can perform a search using an important query provided from the important query extraction device 10 and can obtain information to be sought.

[Operation of important query extractor]
Next, the important query extraction processing in the important query extraction apparatus 10 will be described based on the flowchart of FIG. The process shown in FIG. 3 is implemented by executing the important query extraction program as described above.
The query acquisition unit 11 of the important query extraction device 10 acquires a query transmitted from the terminal device 30 (step S11). The query acquisition unit 11 stores the acquired query in the query log storage unit 17 and outputs it to the node setting unit 12.
The first node setting unit 121 of the node setting unit 12 sets the acquired query as the first node (step S12).

  Next, the second node setting unit 122 extracts a query including the word of the first node from the query log storage unit 17, and sets the second node (step S13). Further, the second node setting unit 122 sets a directed edge between the nodes, creates a directed graph as a model indicating the relationship between the queries, and stores the directed graph in the query relationship storage unit 18 (step S14).

Next, the score setting unit 13 calculates the score of each edge and stores it in the query relation storage unit 18 (step S15).
The query output unit 14 searches for the highest value in the score of each edge, and extracts an important query (step S16).
Further, the query output unit 14 outputs the extracted important query to the terminal device 30 (step S17).
Thus, the important query extraction process ends. The important query extraction device 10 executes the processes of steps S11 to S17 each time a query is input from the terminal device 30.

  In the terminal device 30, the important query output from the important query extraction device 10 is displayed. When the user selects the displayed query, the terminal device 30 outputs the query to the search engine 23, and acquires and displays the search result.

[Effects of Embodiment]
According to the above embodiment, the following effects can be obtained.
The important query extraction device 10 extracts a query having a word included in the query input from the terminal device 30 from the query log storage unit 17 and sets it in each node. Then, the edge score between the nodes is set based on the value indicating the rarity of the word common between the nodes and the number of words of the node in the direction toward the edge, and is a highly rare word, and The score of the edge toward the node with a small number of words was set to be high. For this reason, when a query composed of proper nouns and common nouns is input, if the query of the proper noun alone is stored in the query log storage unit 17, the query can be extracted as an important query and output to the terminal device 30. . Usually, when a word having a high rarity such as a proper noun, that is, a small number stored in the query log is input, it can be estimated that the user most demands information related to the word. Therefore, a query including the word with a small number of words can be determined as the most important query.
For this reason, the user of the terminal device 30 can obtain an important query from the important query extraction device 10 that does not include a spelling error or a misrepresentation, for example, a correct notation of a specific expression, and performs a new search using the important query. Doing so increases the possibility of getting the information you want.

  In the present embodiment, the number of queries stored in the query log storage unit 17 is used as a numerical value representing the scarcity of words. Therefore, the setting of each node and the calculation of the edge score can be performed based only on the query log (history) stored in the query log storage unit 17. That is, in the present embodiment, in order to extract the important query, it is only necessary to store the query history in the query log storage unit 17. For example, the behavior such as how each user inputs a plurality of queries. There is no need to prepare logs with history. For this reason, a system load is small and an important query can be extracted with a simple system.

  Furthermore, even when the input frequency of similar queries including spelling errors is high, in this embodiment, only queries including matching words are set as nodes, and edges are not set for similar words. For this reason, it is possible to prevent a similar query including a spelling error from being erroneously determined as an important query.

[Modification]
In addition, this invention is not limited to embodiment mentioned above, In the range which can achieve the objective of this invention, the deformation | transformation shown below is also included.
For example, in the above embodiment, the information search device 20 and the important query extraction device 10 are configured as separate server devices, but may be configured as a single server device.

  Further, the formula for calculating the score of the edge is not limited to the formula described in the formula (2). In other words, any expression may be used as long as the word rarity of the query is high, the score is high, and the score is high as the number of words included in the query is small.

  INDUSTRIAL APPLICABILITY The present invention can be used as an important query extraction device, an important query extraction method, and an important query extraction program that can determine and output a query with high importance in relation to an input query.

  DESCRIPTION OF SYMBOLS 1 ... Search system, 10 ... Important query extraction apparatus, 11 ... Query acquisition part, 12 ... Node setting part, 121 ... 1st node setting part, 122 ... 2nd node setting part, 13 ... Score setting part, 14 ... Query output , 17 ... Query log storage unit, 18 ... Query relation storage unit, 20 ... Information retrieval device, 30 ... Terminal device.

Claims (5)

A query acquisition unit for acquiring a query including a word as a search keyword;
A query log storage unit for sequentially storing the queries acquired by the query acquisition unit as a query log;
A first node setting unit that sets a query acquired by the query acquisition unit as a first node;
A second node setting unit that extracts a query including a word included in the query of the first node from a query log stored in the query log storage unit, and sets the extracted query as a second node;
A score setting unit that adds a directed edge between the nodes and sets a score of each directed edge;
A query output unit that identifies an important node based on the score set by the score setting unit and outputs a query set to the node as an important query;
The score setting unit sets the score of the directed edge from one node to the other node based on the number of words included in the other node and a numerical value representing the rarity of words common to each node. An important query extraction device characterized by that. The important query extraction device according to claim 1,
The score setting unit sets the highest score of the directed edge toward the node when there is a node composed only of the rarest word among the words included in the input query. ,
The said query output part outputs the query of the node with the highest score of the said directed edge as an important query. The important query extraction apparatus characterized by the above-mentioned. In the important query extraction device according to claim 1 or 2,
The score setting unit has an edge score from node i to node j as H_ {i, j}, the number of words constituting a query set in node j as N_q_j, and the number of queries including word w in the query log storage unit DF (w) is a numerical value indicating DF (w), and a value obtained by adding the reciprocal of DF (w) for the words included in node i and node j is Σ_ {w∈q_i∩q_j} 1 / DF (w), Edge score,
H_ {i, j} = 1 / N_q_j * (Σ_ {w∈q_i∩q_j} 1 / DF (w))
An important query extraction device characterized by An important query extraction method for extracting an important query by a computer,
The computer
A query log storage unit storing a history of queries,
A query acquisition step for acquiring a query including a word as a search keyword;
A first node setting step of setting the query acquired in the query acquisition step as a first node;
A second node setting step of extracting a query including a word included in the query of the first node from a query history stored in the query log storage unit, and setting the extracted query as a second node;
A score setting step of adding a directed edge between the nodes and setting a score of each directed edge;
A query output step of identifying an important node based on the score set in the score setting step, and outputting a query set in the node as an important query;
In the score setting step, the score of the directed edge from one node to the other node is set based on the number of words included in the other node and a numerical value representing the rarity of words common to each node. An important query extraction method characterized by that. On the computer,
A query acquisition step for acquiring a query including a word as a search keyword;
A first node setting step of setting the query acquired in the query acquisition step as a first node;
A query including a word included in the query of the first node is extracted from a query history stored in a query log storage unit in which a query history is stored, and the extracted query is set as a second node. Node setting step;
A score setting step of adding a directed edge between the nodes and setting a score of each directed edge;
An important query extraction program for executing an query output step of identifying an important node based on the score set in the score setting step and outputting a query set in the node as an important query,
In the score setting step, the score of the directed edge from one node to the other node is set based on the number of words included in the other node and a numerical value representing the rarity of words common to each node. An important query extraction program characterized by that.

Images