Online social networking platforms are an important communication medium for cultural events, as they allow exchanging opinions almost in real-time, by publishing messages during the event itself, but also outside of this period. Word embedding has become a popular way to represent and extract information from such messages. In this paper, we propose a preliminary work aiming at assessing the benefits of taking temporal information into account when modeling messages in the context of a cultural event. We perform statistical and visual analyses on two word different representations: one including temporal information (Temporal Embedding), the second ignoring it (Word2Vec approach). Our preliminary results show that the obtained models exhibit some similarities, but also differ significantly in the way they represent certain specific words. More interestingly, the temporal information conveyed by the Temporal Embedding model allows to identify more relevant word associations related to the domain at hand (cultural festivals).
展开▼
