I am trying to implement this paper A Brand-New Look at You: Predicting Brand Personality in Social Media Networks with Machine Learning for labeling Twitter data of brands with a corresponding brand personality. Basically a brand can exhibit certain traits based on Aaker's Personality Traits for brands as per the image below.
On the top most level there are personality dimensions and below them there are traits associated with these dimensions. I am using the aforementioned paper as I do not have access to any labeled data just data from Twitter of brands.
In the paper they used LDA Topic Clusters and Word2Vec word representations along with dictionaries of the brand personality traits to accomplish this task. I am having issues visualizing the exact steps according to the paper.
The steps based on the paper (page 14-15):
- Create refined clusters of posts utilizing LDA and Word2Vec(these clusters describe the various topics being discussed).
- Take each of the Twitter posts and use Word2Vec distance similarities between the Tweets and each of the refined topics(to see which Tweet most accurately represents each of the refined topic clusters)
Then the Tweets were only included in only one refined topic cluster and the Tweets with a 50% similarity in Word2Vec distance moved on to the next step(more steps such as using the brand personality dictionary with this clusters but I need a better understanding of what I am doing here first).
I am stuck at how I am supposed to use LDA and Word2Vec together. So if I cluster the tweets using LDA and Word2Vec then I would need to compare the distance similarity of the topic clusters from the Word2Vec clusters and the Tweets and if the Tweets are similar(threshold of 50%) they would be used in the next step. I am confused at how LDA plays a role here as I do not know where the topic clusters for it are being used in this step.