Emoji Prediction And Hashtag Recommendation For Social Media

Emoji and hashtags are unique symbols created for major social media platforms,quickly adopted into online conversations and widely accepted by social media users.Tasks for emojis and hashtags have shown their popularity and impact on the culture of natural language processing.As new hotspots of language study,emoji prediction and hashtag recommendation for social media have been given increasing attention.Emoji prediction and hashtag recommendation are tasks that extracting textual information representation from input data,and predicting emojis or hashtags which may appear in a tweet or a microblog.Current search on emoji prediction and hashtag recommendation typically treats emoji prediction and hashtag recommendation as two separate tasks,which ignoring the correlation between them.In this paper,we consider task coupling of emoji prediction and hashtag recommendation,for their same input and certain overlap and interdependence on the usage of emoji and hashtags in tweets or microblogs.Therefore,we propose a double-task joint prediction model(JPM)for emoji prediction and hashtag recommendation.The model can process a tweet or microblog to textual information representation to predict emojis and hashtags at the same time.To demonstrate the validity of JPM,we constructed a corpus containing five datasets,including an English dataset from Twitter and four Chinese datasets from Sina Weibo,which are on the field of senior nursing(pension and retirement),estate,car,and NBA(basketball),to verify if JPM shows same ability to predict on data in different languages and fields.Also,we proposed eight single-task prediction models to predict emojis or hashtags separately as comparison groups.In this paper,we prove the excellent performance of JPM,and analyze advantages and disadvantages of every model to provide ideas and learning directions for future.In summary,this paper develops a series of experiments for methods of emoji prediction and hashtag recommendation.Predictions show that JPM can predict better in different fields,on different social media platforms and in different languages.We expound the advantages and disadvantages and forecast future research directions.