October 27, 2023
By William Dean
In this webinar, we will explore the use of a traditional Natural Language Processing technique for modeling weekly calendar data. We will delve into how Latent Dirichlet Allocation can be applied to discretized calendar events, allowing us to tap into the model's probabilistic origins and its connection to Bayesian principles, offering a wide array of potential applications and insights.
Will Dean is a Statistician and Data Scientist with experience in geospatial and user analytics. He is passionate about Bayesian methods and using data visualization to tell a story. He is interested in software design and how it can make data problems easier and more enjoyable to solve.
00:00 Webinar begins
04:00 Presentation begins
04:29 Will's background
05:11 About the talk
05:57 Case study dataset
06:57 Data in mind
07:33 Timestamps provide more info
07:52 Calendar Visualization
09:50 Data Generation Process (First Attempt)
11:26 Data Generation Process (Second Attempt)
12:32 Discrete Approximation
13:19 Data Generation Process (Third Attempt)
13:57 How to people get around?
13:58 Data Generation Process (Fourth Attempt)
17:41 Latent Dirichlet Allocation
18:26 Use what is available
19:51 Define "Vocab"from timestamps
20:23 Aggregate to "documents"
21:42 Learn from "topics"
23:43 "Topic" insights
24:12 Predict and Transform
26:01 Low data support
27:24 Prior Impact
29:27 Next steps for latent-calendar project
30:44 Where it fits into marketing?
33:57 (Q/A) How was the project perceived by stakeholders?
38:10 (Q/A) Is this being used primarily as an insight generating tool?
39:48 (Q/A) Can you explain the connection to CLV modelling ...
44:14 (Q/A) In general more about properties and exploiting them when modelling ...
53:34 (Q/A) Whenever working with Timestamp data is it cyclical ....?
55:40 Webinar ends