Socio-semantic Query Expansion Using Twitter Hashtags

Knowledge-enhanced document embeddings for text classification

“…We used IBk without and with a weighted vote scheme, which gives for each of the nearest neighbors a weight vote equal to 1/(1s), where s is a similarity measure between neighbors. The values of k were 1, 3,5,7,9,11,13,15,17,19,25,35,45,55. We used both Euclidean and cosine distances as proximity measure.…”

“…It latently deals with text semantics since it reduces the noise caused by synonymy and polysemy. Beyond latent semantics, the use of concepts based on external knowledge sources, like WordNet and Wikipedia [15,16,17,18], related concepts obtained from social networks [19], and the application of natural language processing methods, such as named entity recognition, part-of-speech tagging, and semantic role labeling, are other approaches to enrich the text representation [20,21,22,23].…”

Semantic query suggestion using Twitter Entities

“…In a representative approach [5] and given a query, Efron attempts to statistically identify a number of hashtags relevant to the given query, that may be used to expand it and lead to better results. Even in our own previous work [2], we proposed the utilization of hashtags as the main source of information acquisition, by searching the specific query terms within microblog posts under the condition that the former need to appear as hashtags; then, we calculated the most common hashtags that co-occur together with the original query, and, thus, expanded the query with the new hashtags. Last but not least, the observation that microblog posts are created during an actual event and contain comments and/or information directly related to it leads to event detection research efforts [18] based on posts and/or hashtags.…”

“…As case studies in our experiments, we consider the political situations in Egypt and Syria, which have a constant interest for many years worldwide. Taking into account the knowledge we earned from two other previous work of ours [1,2], we initiated the results procedure by using "Egypt", and "Syria" as seeds that correspond to our case studies. Table 1 presents the entities with the top-k% survivability rate between 16 subsequent secondary samplings appeared in January 13, 2014, thus highlighting the entities that reflect a significant trend behavior with respect to the seed term during that day (k ¼10).…”

“…From Table 3 we notice that within an eight-day period and for the cases of Egypt and Syria, the last suggestions were replaced by newer ones at least three times (10th, 11th, 14th and 12th, 13th, 14th of January, respectively). According to our research in query suggestion for web news services, this is the second time where Reuters receives the best comments with respect to trendy proposed suggestions near a user's query (see the evaluation described in [2]). Finally, in an attempt to directly compare if our approach "captures" the new replacements made by Reuters, we checked whether these suggestions appeared in our records.…”