JP5378272B2 - Query feature value calculation method, apparatus, and program - Google Patents

Description

  The present invention relates to a query feature amount calculation method, apparatus, and program, and more particularly, in recommending a search query to a user, for the purpose of improving search efficiency by a recommended query, and to bias a click distribution of a search query. The present invention relates to a query feature value calculation method, apparatus, and program for calculating a feature value of a query in which a desired page of interest is found.

  In recent years, an attempt has been made to analyze a log file indicating a user's search status in a search service and use it to improve the search service. Here, FIG. 7 shows an example of a search query log. In general, the log file includes information on items (date, user ID, query name, click URL, ranking order) as shown in FIG.

  The item “Date” in FIG. 7 is the date when the user clicked the URL. “User ID” is the ID of the user who searched using the search query. “Query name” is a query name used by the user for the search. “Click URL” is the URL of the search result clicked by the user. “URL ranking order” is the search ranking order of the URL clicked by the user.

  These logs are the superiority or inferiority judgment with respect to the URL of the search result by the user, and are used to measure the accuracy of the search result. In addition, by analyzing the connection between a query and a URL, it can be used to cluster queries with the same search intention or to calculate the importance of the query.

  There is a technique for recommending a query that makes it easy for a user to find a desired page using these log files (see, for example, Non-Patent Document 1). In this technology, the relationship between a query and a URL is calculated using a bipartite graph of the relationship between the query and the URL. Here, the same search intention queries are clustered using the property that the same search intention queries access the same URL group. In the clustered query, a query having a high degree of relevance is used as a representative query. When recommending a query to a user, a query having a high degree of relevance is recommended from a group of queries in the clustering to which the query belongs to a query that the user intends to use. With this technology, the user can perform a search with a query with higher search accuracy.

Hongbo Deng, Irwin King, Michael R. Lyu: Entropy-biased Models for Query Representation on the Click Graph, ACM SIGIR, pages 339-346, 2009

  In the above-described conventional technology, a frequently used query tends to be recommended to the user. Therefore, when a query in which a desired page is not found is used by many users, there is a problem that the search efficiency of the user is lowered by recommending the query to the user.

  The present invention has been made in view of the above points, and an object of the present invention is to provide a query feature amount calculation method, apparatus, and program capable of recommending to a user a query in which a page desired by another user is found. .

  FIG. 1 is a diagram for explaining the principle of the present invention.

The present invention (Claim 1) is a query feature amount calculation method for calculating a feature amount of a search query of a search service,
A search query log storage means for storing a click log for a search query of a search service;
URL information storage means for storing information on the URL;
In an apparatus having query feature value calculation means for calculating a query feature value for query recommendation to a user,
The query feature quantity calculation means
When a search query is input, a click log including the same query as the query is searched from the search query log storage unit, and the query is calculated based on a click feature amount calculation rule using a click distribution for URLs in the click log. click feature quantity click feature amount calculation step of calculating (step 1),
PageRank value about the URL of the URL information storage unit, and URL feature amount calculation step of calculating the URL feature amount to the query by applying a URL specificity calculation rule (Step 2),
A query feature amount calculating step (step 3) is performed in which the query feature amount is calculated using the click feature amount and the URL feature amount.

  Further, according to the present invention (Claim 2), in the query feature amount calculation method according to Claim 1, for a click log including the same query, a numerical value obtained by calculating the entropy of the URL included in the click log, and the click log A click feature amount calculation rule is used that is calculated by multiplying the average value of the search ranking rank of the included URL.

  Further, according to the present invention (Claim 3), in the query feature value calculation method according to Claim 1, the page rank value of each URL is changed to the page rank value of the URL included in the click log including the same query. A URL specificity calculation rule that calculates entropy for all URLs in the click log including the same query using a value divided by the sum of rank values is used.

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

The present invention (Claim 4) is a query feature amount calculation device for calculating a feature amount of a search query of a search service,
Search query log storage means 5 storing a click log for a search query of a search service;
URL information storage means 6 for storing information on the URL;
A click log including the same query as the input query is searched from the search query log storage unit 5, and the click feature amount of the query is calculated based on the click feature amount calculation rule using the click distribution with respect to the URL in the click log. and click feature calculating unit 2 for calculating,
PageRank value about the URL of the URL information storage means 6, the URL feature calculating unit 3 for calculating the URL feature amount to the query by applying a URL specificity calculation rule,
And a query feature amount calculation unit 1 that calculates a query feature amount using a click feature amount and a URL feature amount.

Further, the present invention (Claim 5) defines the click feature amount calculation rule of Claim 4 as follows:
For a click log that includes the same query, it is calculated by multiplying the numerical value obtained by calculating the entropy of the URL included in the click log and the average value of the search ranking rank of the URL included in the click log.

Further, according to the present invention (Claim 6), the URL specificity calculation rule of Claim 4 is:
All URLs in the click log that contain the same query using the value obtained by dividing the page rank value of each URL by the sum of the page rank values of the unique URLs for the page rank value of the URLs contained in the click log that contains the same query The entropy for the target is calculated.

  The present invention (Claim 7) is a query feature quantity calculation program for causing a computer to function as each means constituting the query feature quantity calculation apparatus according to any one of claims 4 to 6.

  While the query feature recommended in the prior art is a query with a high user usage rate, as described above, in the present invention, the bias distribution of the click distribution of the search query log is calculated as entropy as the query feature amount. To do. Furthermore, in order to express whether the URL that the user wants is included in the URL group of the search result as a feature amount, the entropy for the page rank value of the URL group is calculated, and the query feature amount is calculated by combining these two entropies. calculate. Thereby, it is possible to recommend a query that is highly likely to obtain a good search result for the user.

It is a figure for demonstrating the principle of this invention. It is a principle block diagram of this invention. It is a block diagram of the query feature-value calculation apparatus in one embodiment of this invention. It is a flowchart of the query feature-value calculation in one embodiment of this invention. It is an example of data of the search query log memory | storage part in one embodiment of this invention. It is a data example of the URL information storage part in one embodiment of this invention. It is an example of a search query log.

  Hereinafter, embodiments of the present invention will be described with reference to the drawings.

  In the present invention, the bias distribution of the click distribution of the search query log is calculated as entropy as the query feature amount. Furthermore, in order to express whether the URL that the user wants is included in the URL group of the search result as a feature amount, the entropy for the page rank value of the URL group is calculated, and the query feature amount is calculated by combining these two entropies. calculate.

  FIG. 3 shows a configuration of the query feature quantity calculation apparatus according to the embodiment of the present invention.

  The query feature quantity calculation device includes a query feature quantity calculation unit 1, a click entropy calculation unit 2, a page rank entropy calculation unit 3, a query extraction unit 4, a search query log storage unit 5, and a URL information storage unit 6.

  When the query is input, the query feature quantity calculation unit 1 transmits a query name as a calculation request to the click entropy calculation unit 2 and the page rank entropy calculation unit 3, receives the click entropy from the click entropy calculation unit 2, The page rank entropy is acquired from the page entropy calculation unit 3 and the query feature amount is calculated.

  When the click entropy calculation unit 2 acquires the query name from the query feature amount calculation unit 1, the click entropy calculation unit 2 transmits a search query log collection command for the query name to the query extraction unit 4.

  The query extraction unit 4 extracts the search query log corresponding to the query name from the search query log storage unit 5 based on the search query log collection command acquired from the click entropy calculation unit 2 and returns it to the click entropy calculation unit 3. In addition, the click URL is extracted from the search query log storage unit 5 based on the query acquired from the page rank entropy calculation unit 3, the URL information storage unit 6 is searched based on the URL, and the page rank value related to the URL is extracted, Return to page rank entropy calculation unit 5.

  When the click entropy calculation unit 2 acquires the search query log related to the query name from the query extraction unit 4, the click entropy calculation unit 2 calculates the click entropy based on a predetermined rule and returns it to the query feature amount calculation unit 1.

  When the page entropy calculation unit 3 acquires the query name from the query feature quantity calculation unit 1, the page entropy calculation unit 3 transmits the query name to the query extraction unit 4, acquires a search query log corresponding to the query name and a page rank value related to the URL, The page rank entropy is calculated based on the page rank value based on a predetermined rule, and transmitted to the query feature quantity calculation unit 1.

  The search query log storage unit 5 stores data of items similar to those in FIG.

  The URL information storage unit 6 stores data of items of URL, page rank value, and title.

  Next, the operation in the above configuration will be described.

  FIG. 4 is a flowchart of query feature value calculation according to an embodiment of the present invention.

  Step 101) When a query is input, the query feature quantity calculation unit 1 sends the query name to the click entropy calculation unit 2. Next, the click entropy calculation unit 2 sends the query name to the query extraction unit 4, and the query extraction unit 4 searches the search query log storage unit 5 with the query name and returns the click log to the click entropy calculation unit 2. To do.

  Step 102) The click entropy calculation unit 2 calculates the number of appearances for each URL for the acquired click log.

  Step 103) The click entropy calculation unit 2 calculates the click entropy by the following formula (click feature amount calculation rule).

Here, P (u) is the number of occurrences of the URL in the log of the same query including the query,
P (u) = total number of URLu / query q
It is calculated by the total number of URL appearances in the log. The average ranking is the average ranking of URL rankings in the query q log.

ClickEntropy (q) = Average ranking rank x (-1.0 x Σ (P (u) x log (P (u)))
The obtained click entropy is returned to the query feature quantity calculation unit 1 together with the query log.

  Step 104) The query feature quantity calculation unit 1 extracts the rank value (ranking rank) of each URL page included in the click log. The query is transmitted to the page rank entropy calculation unit 3. The page rank entropy calculation unit 3 transmits the query to the query extraction unit 4. The query extraction unit 4 extracts the click URL corresponding to the query from the search query log storage unit 5 based on the query, and then extracts the page rank value of each URL from the URL information storage unit 6 based on the click URL. To the page rank entropy calculation unit 3.

  When the page rank entropy calculation unit 3 receives the page rank value related to the URL, the page rank entropy calculation unit 3 calculates the page rank entropy by the following formula (URL specificity calculation rule). Here, Pagerank (u) is the ratio of the page rank value to the unique URL (URLu) in the log of the same query including the query. Pagerank (u) = Page rank value of URLu / Total page rank value of unique URL Is calculated by

PagerankEntropy (q) = − 1.0 × Σ {Pagerank (u) × log (Pagerank (u))}
The calculated page rank entropy is transmitted to the query feature quantity calculation unit 1.

  Step 105) With respect to the entropy calculated in Steps 103 and 104, the query feature quantity calculation unit 1 uses a value obtained by multiplying the click entropy and the page rank entropy by the following formula as a query feature quantity.

Query q feature = ClickEntropy (q) x PagerankEntropy
The smaller QueryEntropy (q) above, the more satisfying the URL is obtained for the user who used the query q. The smaller the PagerankEntropy (q), the easier it is to specify the search result for the query q ( This indicates that the desired page has been found, including a corporate website and a clear purge of interest.

  The above specific examples are shown below.

  In the following, it is assumed that a plurality of search query logs as shown in FIG. 5 are stored in the search query log storage unit 5, and the URL information storage unit 6 includes URL names and search query logs for a plurality of URLs. It is assumed that the page rank value and title of the URL are stored.

  Hereinafter, description will be made along the flowchart of FIG.

  Step 101) When a query is input to the query feature quantity calculation unit 1, the query feature quantity calculation unit 1 transmits a query name as a calculation request to the click entropy calculation unit 2. Here, it is assumed that “misuho” is input as the query name.

  When the click entropy calculation unit 2 receives “misuho” as a query from the query feature amount calculation unit 1, the click entropy calculation unit 2 transmits a search query log collection command to the query extraction unit 4.

  When the query extraction unit 4 receives a query from the click entropy calculation unit 1, the query extraction unit 4 extracts a search query log having “misuho” as a query name from the search query log storage unit 5. Here, five click logs are extracted from the search query log storage unit 5 shown in FIG. Then, the extracted search query log is transmitted to the click entropy calculation unit 2.

  Step 102) Upon receiving the search query log related to “misuho” from the query extraction unit 4, the click entropy calculation unit 2 extracts the URL ranking (1, 4, 10, 3, 7), and calculates the click entropy by the following formula. calculate.

ClickEntropy (q) = Average ranking rank x (-1.0 x Σ (P (u) x log (P (u)))
= (1 + 4 + 10 + 3 + 7) / 5 × (−1.0)) ×
(1/5 × log1 / 5 + 1/5 × log1 / 5 +
(1/5 × log1 / 5 + 1/5 × log1 / 5 + 1/5 × log1 / 5)
= 25/5 × (−1.0) × (0.2 × (−0.698) + 0.2 × (−0.698) +
0.2 × (−0.698) + 0.2 × (−0.698) +
0.2 × (−0.698))
= 5 × (−1.0) × (−0.698)
= 3.494
The click entropy value calculated above is transmitted to the query feature quantity calculation unit 1.

  Step 103) Upon receiving the query name “misuho” from the query feature quantity calculation unit 1, the page rank entropy calculation unit 3 transmits the query name “misuho” to the query extraction unit 4.

  When the query extraction unit 4 receives the query name “misuho” from the page rank entropy calculation unit 3, the query extraction unit 4 extracts a search query log having the query name “misuho” from the search query log storage unit 5. In this example, five click logs are extracted from the search log storage unit 5 shown in FIG.

  Next, for the five URLs included in the extracted click log, page rank values relating to the five URLs are extracted from the URL information storage unit 6 and transmitted to the page rank entropy calculation unit 3.

  Step 104) When the page rank entropy calculation unit 3 receives the page rank value related to the URL from the query extraction unit 4, the page rank entropy calculation unit 3 calculates the page rank entropy by the following formula. Here, the sum of the page rank values for the five URLs is 0.4 + 0.3 + 0.3 + 0.2 + 0.1 = 1.3.

PagerankEntropy (q) = -1.0 × Σ {Pagerank (u) × log (Pagerank (u))}
= −1.0 × (0.4 / 1.3 × (−0.512) + 0.3 / 1.3 × (−0.637) +
0.3 / 1.3 × (−0.637) + 0.2 / 1.3 × (−0.813) +
0.1 / 1.3 × (−1.114)
= −1.0 × ((− 0.158) + (−0.147) + (−0.147) + (−0.125) +
(‐0.086))
= 0.663
The page rank entropy value calculated above is transmitted to the query feature quantity calculation unit 1.

  Step 105) When the query feature quantity calculation unit 1 receives the click entropy value and the page rank entropy value from the click entropy calculation unit 2 and the page rank entropy calculation unit 3, respectively, the query feature amount calculation unit 1 uses the following formula to change the query name “misuho”. A query feature amount is calculated.

Query feature (misuho) = 3.494 × 0.663
= 2.317
Next, when “Misho Bank” is input as a query name to the query feature quantity calculation unit 1, the same query calculation quantity as described above is obtained.

ClickEntropy (q) = Average ranking × (−1.0 × Σ (P (u) × log (P (u))))
= (1 + 1 + 2 + 1 + 1 + 2) / 6 × (−1.0) × (4/6 × (−0.176) +
2/6 x (-0.477))
= 1.33 × (−1.0) × ((−0.117) + (−0.159))
= 1.33 × 0.276
= 0.367
PagerankEntropy (q) = −1.0 × Σ {Pagerank (u) × log (Pagerank (u))}
= −1.0 × (0.9 / 1.2 × (−0.125) + 0.3 / 1.2 × (−0.602))
= −1.0 × ((−0.094) + (−0.150))
= -1.0 x 0.244
= 0.244
Query feature (Misho Bank) = 0.367 × 0.244
= 0.090
From the above query characteristics, since the value of “Misuho Bank” is smaller than the query name “misuho” (because it is entropy), the query name “Misuho Bank” is likely to be recommended.

  Each operation of the constituent elements of the query feature quantity calculation device shown in FIG. 3 described above is constructed as a program, installed in a computer used as the query feature quantity calculation device, executed, or distributed via a network. Is possible.

  Further, the constructed program can be stored in a portable storage medium such as a hard disk, a flexible disk, or a CD-ROM, and can be installed or distributed in a computer.

  The present invention is not limited to the above-described embodiment, and various modifications and applications can be made within the scope of the claims.

DESCRIPTION OF SYMBOLS 1 Query feature-value calculation means, Query feature-value calculation part 2 Click feature-value calculation means, Click entropy calculation part 3 URL feature-value calculation means, Page rank entropy calculation part 4 Query extraction part 5 Search query log storage means, Search query log storage part 6 URL information storage means, URL information storage unit

Claims (7)

A query feature amount calculation method for calculating a feature amount of a search query of a search service,
Search query log storage means for storing a click log for the search query of the search service;
URL information storage means for storing information on the URL;
In an apparatus having query feature value calculation means for calculating a query feature value for query recommendation to a user,
The query feature amount calculating means includes:
When a search query is input, a click log including the same query as the query is searched from the search query log storage unit, and a click distribution for a URL in the click log is used to calculate the click feature amount rule. and click feature amount calculation step of calculating the click feature amount of the query,
PageRank value related to the URL of the URL information storage unit, and URL feature amount calculating step of applying a URL specificity calculating rule calculates the URL feature amount with respect to the query,
A query feature amount calculating step of calculating a query feature amount using the click feature amount and the URL feature amount;
A query feature amount calculation method characterized by: The content of the click feature amount calculation rule is as follows:
The query feature according to claim 1, wherein a click log including the same query is calculated by multiplying a numerical value obtained by calculating an entropy of a URL included in the click log and an average value of search rankings of URLs included in the click log. Quantity calculation method. The contents of the URL specificity calculation rule are as follows:
All URLs in the click log that contain the same query using the value obtained by dividing the page rank value of each URL by the sum of the page rank values of the unique URLs for the page rank value of the URLs contained in the click log that contains the same query The query feature amount calculation method according to claim 1, wherein entropy for the target is calculated. A query feature amount calculation device for calculating a feature amount of a search query of a search service,
Search query log storage means for storing a click log for the search query of the search service;
URL information storage means for storing information on the URL;
A click log including the same query as the input query is searched from the search query log storage means, and the click feature amount of the query is calculated based on the click feature amount calculation rule using the click distribution for the URL in the click log. and click feature quantity calculating means for calculating,
PageRank value related to the URL of the URL information storage unit, and URL feature quantity calculating means for calculating the URL feature amount with respect to the query by applying a URL specificity calculation rule,
Query feature value calculating means for calculating a query feature value using the click feature value and the URL feature value;
A query feature quantity calculation device characterized by comprising: The click feature amount calculation rule is:
The query feature according to claim 4, wherein the click log including the same query is calculated by multiplying the numerical value obtained by calculating the entropy of the URL included in the click log and the average value of the search ranking rank of the URL included in the click log. Quantity calculation device. The URL specificity calculation rule is:
All URLs in the click log that contain the same query using the value obtained by dividing the page rank value of each URL by the sum of the page rank values of the unique URLs for the page rank value of the URLs contained in the click log that contains the same query The query feature amount calculation apparatus according to claim 4, wherein entropy is calculated for the target.   A query feature value calculation program for causing a computer to function as each means constituting the query feature value calculation device according to any one of claims 4 to 6.

Images