US 2011/0213761 A1 · SEARCHABLE WEB SITE DISCOVERY AND RECOMMENDATION

Abstract

A searchable web site discovery and recommendation system and method for discovering and recommending searchable web sites to a user of a search engine. In general, given a search query by a user embodiments of the system and method discover and then recommend a list of searchable web sites. Embodiments of the system and method discover and recommend searchable web sites using client-side query behavior logs. A set of descriptive features is automatically extracted from the logs. In addition, embodiments of the system and method expand the initial search query by submitting the query to a general search engine and leveraging snippets from the search engine results. Embodiments of the system and method model a conditional probability of a searchable web site based on four components and the extracted descriptive features. A ranked list of recommended searchable web sites is obtained from the conditional probability and displayed to a user.

Claims20

Dependent claims

A method implemented on a computing device having a processor for recommending searchable web sites to a user, comprising:

using the computing device having the processor to perform the following:
obtaining query behavior using client-side query behavior logs;
discovering searchable web sites using the client-side query behavior logs;
inputting a search query from the user;
submitting the search query to a general search engine to obtain a query language model;
computing a probability of each searchable web site given the search query by using the query language model;
ranking each searchable web site based on the computed probability;
creating a list of top-ranked searchable web sites based on rank to obtain a ranked list of recommended searchable web sites;
and
displaying the ranked list of recommended searchable web sites to the user.

7 more elements

A method implemented on a computing device having a processor for discovering and recommending searchable web sites, comprising:

using the computing device having the processor to perform the following:
discovering searchable web sites using client-side query behavior logs that record activities of users within a browser to generate a list of discovered searchable web sites;
inputting a search query from a user;
extracting data from the client-side query behavior logs to obtain descriptive features for each of the searchable web sites on the list of discovered searchable web sites;
submitting the search query to a general search engine;
obtaining snippets of search engine result pages from the general search engine;
expanding the search query using the snippets from top results of the search engine result pages to obtain an expanded search query;
generating a query language model for the expanded search query as a multinomial distribution;
computing a probability of each of the searchable web sites using the query language model;
ranking each of the searchable web sites based on the compute probability;
and
generating a ranked list of recommended searchable web sites based on the ranking.

9 more elements

A computer-implemented method for discovering and recommending searchable web sites to a user, comprising:

obtaining client-side browsing logs that record activities of user within a browser;
discovering searchable web sites using the client-side browsing logs to obtain a list of discovered searchable web sites;
inputting a search query from the user;
modeling a searchable web site as a conditional probability of a searchable web, v, given the search query, q, by using the equation:
P

(
v
|
q
)
=

∑
w
∈
Vocab

P

(
w
,
v
|
q
)
=

∑
w
∈
Vocab

P

(
w
|
q
)

P

(
v
|
q
,
w
)
=

∑
w
∈
Vocab

P

(
w
|
q
)

P

(
v
|
w
)
=

∑
w
∈
Vocab

P

(
w
|
q
)

P

(
w
|
v
)

1
P

(
w
)

P

(
v
)
,
where, w, represents a term in a vocabulary, which is denoted as Vocab, P(w/q) is a conditional probability of the term w given the query q, P(w/v) is a conditional probability of the term w given the searchable site v, 1/P(w) is the inverse of the probability of the term w, and P(v) is a static rank of the searchable site v;
ranking each searchable web site based on the conditional probability;
generating a ranked list of recommended searchable web sites based on the ranking;
and
displaying to the user the ranked list of recommended searchable web sites along with regular search results from a general search engine.

111 more elements

Dependencies are inferred from the claim text.

Searchable web site discovery and recommendation

Life of this patent

Searchable web site discovery and recommendation

Life of this patent

Abstract

Claims20