Understand your Brand

Insight IDX presentation

Transcript

1. Understand Your Brand Eric Titus
2. Your Brand “Accelerate marketing and product through social”
3. Extract Relevant Topics ⚬ Commercial
4. Extract Relevant Topics ⚬ Commercial ⚬ Social Fun ⚬ ...and more (~300 topics)
5. Automate ⚬ no sampling! ⚬ fast!
6. Topics? Automate ⚬ no sampling! ⚬ fast! ⚬ ~ 2 million posts ⚬ in “Russian” ⚬ product: “Budweiser”
7. Workflow NLTK custom filtering LDA Optimize topics ⚬ Keywords – What is this topic? ⚬ Probable topics in each post – What topic does a post belong to? extract usable text Identify, count, remove retweets
8. ⚬ “Россия” ⚬ “Москва” ⚬ Рыба Translate(keywords)→ Topics
9. ⚬ “Russia” ⚬ “Moscow” ⚬ Fish 1. National Identity Translate(keywords)→ Topics
10. ⚬ “Russia” ⚬ “Moscow” ⚬ Fish 1. National Identity ⚬ Required ⚬ ❤ ⚬ Drinking 2. Brand Loyalty Translate(keywords)→ Topics
11. ⚬ “Russia” ⚬ “Moscow” ⚬ Fish 1. National Identity ⚬ Required ⚬ ❤ ⚬ Drinking 2. Brand Loyalty 3. Debate on national beverage Translate(keywords)→ Topics ⚬ “Tea” ⚬ Who knows ⚬ Who says
12. ⚬ “Russia” ⚬ “Moscow” ⚬ Fish 1. National Identity ⚬ Required ⚬ ❤ ⚬ Drinking 2. Brand Loyalty 3. Debate on national beverage ⚬ Topic labels are automatically applied ⚬ Topics extracted from a few keywords ⚬ Can easily be used with any language Translate(keywords)→ Topics ⚬ “Tea” ⚬ Who knows ⚬ Who says
13. Eric Titus PhD in Chemistry University of Texas at Austin Postdoc at Temple University Super-resolution imaging of single molecules and nanoparticles
15. Minimum shows the optimum number of topics to use in LDA to characterize this data set Unique posts (not retweets) are relatively evenly distributed across topics (top). When looking at all data, including retweets, we see that certain categories are retweeted more than others (bottom). topics are given as numbers to protect brand identity
16. All twitter activity for the brand increases on Saturdays, for both unique content and retweeted content We see unique postings occur primarily in the evening