Social media has emerged to be an important source of information. Entity linking in social media provides an effective way to extract useful information from microposts shared by the users. Entity linking in microposts is a difficult task as they lack sufficient context to disambiguate the entity mentions. In this paper, we do entity linking by first identifying entity mentions and then disambiguating the mentions based on three different features: (1) similarity between the mention and the corresponding Wikipedia entity pages; (2) similarity between the mention and the tweet text with the anchor text strings across multiple webpages, and (3) popularity of the entity on Twitter at the time of disambiguation. The system is tested on the manually annotated dataset provided by Named Entity Extraction and Linking (NEEL) Challenge 2014, and the obtained results are on par with the state-of-the-art methods.
Wikipedia based measure Twitter popularity based measure
Social media networks like Twitter have emerged to be major platforms for sharing information in form of short messages (tweets). Analysis of tweets can be useful for various applications like ecommerce, entertainment, recommendations, etc. Entity linking is the one such analysis task which deals with finding correct referent entities in the knowledge base for various mentions in the tweet. Entity linking in social media is important as it helps in detecting, understanding and tracking information about an entity shared across social media.
Copyright c 2014 held by author(s)/owner(s); copying permitted only for private and academic purposes.
Published as part of the #Microposts2014 Workshop proceedings, Cavoapiylaribglhetoisnlhienled absy CthEeUauRthVoro/lo-w11n4e1r(s()h.ttp://ceur-ws.org/Vol-1141) 2. 2.1
Mention detection is the task of finding entity mentions in the
given text. We assumed mentions as named entities present
inside the tweets. Various approaches for named entity recognition
in tweets have been proposed recently [
Entity disambiguation is the task of assigning the correct referent
entity from the knowledge base to the given mention. We
disambiguate the entity mention using three measures as described below.
The scores from these three measures are combined using
LambdaMART [
This measure disambiguates a mention by calculating the
frequency of occurrence of the mention in the Wikipedia corpus. Wikipedia’s
context based measure has been used in various approaches for
disambiguating mentions in tweets [
Google Cross-Wiki Dictionary (GCD) [
Tweets about entities follow a bursty pattern. Bursty patterns are the bursts of tweets that appear after an event relating to an entity happens. We exploited this fact and tried to measure the number of times the given mention refers to a particular entity on Twitter recently. The mention is queried on Twitter API2 and the resultant tweets are analyzed. All the tweets along with the mention are then queried on the GCD and the candidate entities are taken. Based on the scores returned using GCD, all the candidate entities are ranked (which is the ranked list using M3). As Twitter popularity based measure captures the people’s interests at a particular time, it works well for entity disambiguation on recent tweets. In essence, the methods M2 and M3 are similar but with different inputs. Both use GCD, and produce candidate mentions and score as output. However, M2 takes mention and single tweet text as input whereas M3 takes mention and multiple tweets as input.
We have three rankings available using M1, M2, M3. Now the task is to arrive at the final ranking of the candidate entities by combining the rankings of the three different models. The rankings of different models should be combined such that the overall F1 score is maximized. For this, we use LambdaMART which combines LambdaRank and MART models. LambdaMART creates boosted regression trees for combining the rankings of the three different systems.
The dataset comprises of 2.3K tweets each annotated with the entity mention and its corresponding DBpedia URL. We divided the dataset into the 7:3 (train:test) ratio. Table 1 shows the results obtained using the NEEL Challenge evaluation framework. The best results are obtained when a combination of all the measures were used for disambiguation3. A 5-fold cross validation on the dataset gave an average F1 of 0.52 for M1+M2+M3. 4.
For effective entity linking, mention detection in tweets is important. We improve the accuracy of detecting mentions by combining various Twitter POS taggers. We resolve multiple mentions, abbreviations and spell variations of a named entity using the Google Cross-Wiki Dictionary. We also use popularity of an entity on Twitter for improving the disambiguation. Our system performed well with a F1 score of 0.512 on the given dataset.