This Podcast is part of a series of a student's project about Big Data Businesses and Projects at EM Lyon Business School.

Interviewers (in order of appearance): 
- Ariel W.
- Tobias G.
- Milo H.
Interviewees (from www.upfluence.com): 
- Nicolas Miachon
- Thomas Gallice

Questions and Timings:
Q1: Could you please tell us something about yourselves – your background and maybe about your responsibilities and tasks that you follow at Upfluence. (Starts: 00:15)
Q2: Upfluence is a premier software platform for influencer marketing at scale. Well-known companies including PayPal, Microsoft, Nestlé, Air France and many more, are already highly content customers of your product. Could you describe your core business in a few sentences? (Starts: 02:28)
Q3: The problem you are trying to solve for your clients could look like the following: a client wants to promote his new product or service to the appropriate target audience. However, the customer does not know how to effectively proceed in this situation, so they contact Upfluence. Together, you create a tailored influencer marketing strategy. What does this strategy consist in? And what are the next steps after the situation is clear and the best strategy has been chosen? (Starts: 07:17)
Q4: After looking at the customer-facing side of Upfluence – how you help your clients to solve their problems – we want to understand what happens behind the scene. In one of our previous conversations, you explained to us that your whole process mainly consists of three steps: 1) Data gathering, 2) Classification or categorization and 3) Data aggregation. Could you briefly explain what happens in each of these steps, before we dive into each individual step? (Starts: 10:45)
Q5: In the data gathering phase, you mainly use two techniques to get your data: when available, you use API’s (e.g. Facebook, Twitter and Instagram), and for gathering information from blogs and other relevant sources like websites, you use a crawler. Could you explain this process in more detail? (Starts: 13:15)
Q6: How are you able to distinguish between what is relevant (Influencer-) data you wish to gather from irrelevant data? (Starts: 14:35)
Q7: The second step is categorization. In this step, you use rule-based text mining techniques, and other techniques, such as Bag-of-Words and word vectors (word2vec), to categorize the influencers (e.g. by their niche, location, preferences, audience, etc.). Could you provide an illustrative example on what exactly happens in this step? (Starts: 16:40)
Q8: You are working on leveraging the power of more machine learning techniques in this step, such as using Convolutional Neural Networks (CNNs) to enable e.g. image recognition. Could you explain what additional value you are expecting from the employment of these techniques? (Starts: 20:10)
Q9: The last stage in the three-step-process is aggregation. This is where all the gathered and categorized data comes together from the different platforms, such as Twitter, Facebook, Instagram and LinkedIn. You have advanced your models so far, that you are able to identify if two profiles on different platforms belong to the same person or influencer. Could you explain the basic idea on how you achieve this? (Starts: 23:04)
Q10: The portfolio of Upfluence counts more than 1.3M influencers, over 500M indexed content pieces and you perform over 8.7M updates on a daily basis, to ensure that your data is always up to date. Moreover, you deliver answers to complex queries in your database in only a few seconds. To deal with this amount of data at such high speeds, you need a special type of database architecture, which enables a high degree of parallelized operations. Could you tell us something about this architecture? (Starts: 25:33)
Q11: As a last topic, it would be great if you could summarize what we have talked about, by using the real-world example of the Dagobear sampling campaign. (Starts: 28:57)
We hope that you enjoyed the podcast.

Milo RH