SYSTEM AND METHOD FOR EXPLOITING USER FEEDBACK TO DERIVE TRENDING TERMS AND APPLICATIONS THEREOF
The present teaching relates to trending term identification. Information in different categories from different sources associated with each of terms being evaluated for trendiness is obtained within a recent period and linked to generate a data group for each term. Features are extracted for each term based on information in the corresponding data group and are used to compute a trendiness score in accordance with a scoring model. Trending terms are selected from the terms based on their trendiness scores.
1 . A method, comprising:
with respect to each of multiple term candidates being evaluated on trendiness,
obtaining information associated with the term candidate in different categories from different sources, wherein the information satisfies a recency requirement defined based on a time period,
linking the term candidate and the information associated therewith to generate a data group for the term candidate,
determining features related to the term candidate based on the data group for the term candidate, and
computing a trendiness score for the term candidate based on the features in accordance with a scoring model; and
selecting trending terms from the multiple term candidates based on the trendiness scores of the multiple term candidates.
2 . The method of claim 1 , wherein the different categories include:
search logs each of which records searches conducted using different terms;
content covering at least one topic consistent with what the term candidate represents; and
engagement data with respect to the content.
3 . The method of claim 2 , wherein the engagement data includes one or more of:
click through rate (CTR);
conversion rate (CVR); and
dwell time.
4 . The method of claim 1 , wherein the different sources include:
public online sources including at least one of a content portal, a search engine, and a website;
semi-private sources including at least one of a social media platform, a membership-based interest group, and a chatroom; and
private sources including at least one of electronic mails and communication messages.
5 . The method of claim 1 , wherein the features related to the term candidate includes at least one of:
a category classification for the term candidate;
information on historic popularity of the term candidate;
freshness of the term candidate; and
one or more statistics computed based on the information included in the data group of the term candidate.
6 . The method of claim 5 , wherein the one or more statistics include:
a first metric representing the number of searches conducted during the time period using the term candidate;
a second metric representing the number of articles in the content relating to the term candidate that are available during the time period;
a third metric characterizing the number of occurrences of the term candidate in the articles;
a fourth metric on the number of clicks on content items included in the content; and
a fifth metric indicative of engagement with each content item in the content related to the term candidate accessed during the time period.
7 . The method of claim 1 , wherein
the scoring model specifies a function of the features related to the term candidate;
the step of selecting trending terms from the multiple term candidates comprises:
ranking the multiple term candidates to generate a ranked list of term candidates according to the respective trendiness scores of the multiple term candidates, and
determining the trending terms from the ranked list based on the ranks of the multiple term candidates, wherein
data groups associated with the determined trending terms are labeled as positive data to generate labeled training data for training the scoring model via supervised learning.
8 . A machine readable and non-transitory medium having information recorded thereon, wherein the medium, when read by the machine, causes the machine to perform the following steps:
with respect to each of multiple term candidates being evaluated on trendiness,
obtaining information associated with the term candidate in different categories from different sources, wherein the information satisfies a recency requirement defined based on a time period,
linking the term candidate and the information associated therewith to generate a data group for the term candidate,
determining features related to the term candidate based on the data group for the term candidate, and
computing a trendiness score for the term candidate based on the features in accordance with a scoring model; and
selecting trending terms from the multiple term candidates based on the trendiness scores of the multiple term candidates.
9 . The medium of claim 8 , wherein the different categories include:
search logs each of which records searches conducted using different terms;
content covering at least one topic consistent with what the term candidate represents; and
engagement data with respect to the content.
10 . The medium of claim 9 , wherein the engagement data includes one or more of:
click through rate (CTR);
conversion rate (CVR); and
dwell time.
11 . The medium of claim 8 , wherein the different sources include:
public online sources including at least one of a content portal, a search engine, and a website;
semi-private sources including at least one of a social media platform, a membership-based interest group, and a chatroom; and
private sources including at least one of electronic mails and communication messages.
12 . The medium of claim 8 , wherein the features related to the term candidate includes at least one of:
a category classification for the term candidate;
information on historic popularity of the term candidate;
freshness of the term candidate; and
one or more statistics computed based on the information included in the data group of the term candidate.
13 . The medium of claim 12 , wherein the one or more statistics include:
a first metric representing the number of searches conducted during the time period using the term candidate;
a second metric representing the number of articles in the content relating to the term candidate that are available during the time period;
a third metric characterizing the number of occurrences of the term candidate in the articles;
a fourth metric on the number of clicks on content items included in the content; and
a fifth metric indicative of engagement with each content item in the content related to the term candidate accessed during the time period.
14 . The medium of claim 8 , wherein
the scoring model specifies a function of the features related to the term candidate;
the step of selecting trending terms from the multiple term candidates comprises:
ranking the multiple term candidates to generate a ranked list of term candidates according to the respective trendiness scores of the multiple term candidates, and
determining the trending terms from the ranked list based on the ranks of the multiple term candidates, wherein
data groups associated with the determined trending terms are labeled as positive data to generate labeled training data for training the scoring model via supervised learning.
15 . A system, comprising:
a trending term determiner implemented by a processor and configured for, with respect to each of multiple term candidates being evaluated on trendiness,
obtaining information associated with the term candidate in different categories from different sources, wherein the information satisfies a recency requirement defined based on a time period,
linking the term candidate and the information associated therewith to generate a data group for the term candidate,
determining features related to the term candidate based on the data group for the term candidate, and
computing a trendiness score for the term candidate based on the features in accordance with a scoring model; and
selecting trending terms from the multiple term candidates based on the trendiness scores of the multiple term candidates.
16 . The system of claim 15 , wherein the different categories include:
search logs each of which records searches conducted using different terms;
content covering at least one topic consistent with what the term candidate represents; and
engagement data with respect to the content, including one or more of
click through rate (CTR),
conversion rate (CVR), and
dwell time.
17 . The system of claim 15 , wherein the different sources include:
public online sources including at least one of a content portal, a search engine, and a website;
semi-private sources including at least one of a social media platform, a membership-based interest group, and a chatroom; and
private sources including at least one of electronic mails and communication messages.
18 . The system of claim 15 , wherein the features related to the term candidate includes at least one of:
a category classification for the term candidate;
information on historic popularity of the term candidate;
freshness of the term candidate; and
one or more statistics computed based on the information included in the data group of the term candidate.
19 . The system of claim 18 , wherein the one or more statistics include:
a first metric representing the number of searches conducted during the time period using the term candidate;
a second metric representing the number of articles in the content relating to the term candidate that are available during the time period;
a third metric characterizing the number of occurrences of the term candidate in the articles;
a fourth metric on the number of clicks on content items included in the content; and
a fifth metric indicative of engagement with each content item in the content related to the term candidate accessed during the time period.
20 . The system of claim 15 , wherein
the scoring model specifies a function of the features related to the term candidate;
the step of selecting trending terms from the multiple term candidates comprises:
ranking the multiple term candidates to generate a ranked list of term candidates according to the respective trendiness scores of the multiple term candidates, and
determining the trending terms from the ranked list based on the ranks of the multiple term candidates, wherein
data groups associated with the determined trending terms are labeled as positive data to generate labeled training data for training the scoring model via supervised learning.