|
"Audience Interaction Tracking During Live Streaming"
In recent years, the rise of Tiktok short videos, short videos and live broadcasts have become the strongest e-commerce models. Quickly accumulate fans through lively and creative content, and create more interaction, marketing and sales with live broadcasts.
The problem encountered by the current live broadcast mode is how to understand the feedback content of the audience when there are more than 10,000 people online at the same time and a large number of fans interact in the live broadcast.
In addition to increasing the visibility of messages through sponsorship, I also hope to use natural language analysis to understand the audience's thoughts and conduct long-term tracking to increase the conversion rate.
This project uses Youtube Live Chat as an example to capture the interactive content of the chat room audience during the live broadcast
Used Google Cloud Platform, YouTube Data API and crawlers to capture channel subscription status, video views, audience appearance and chat room content. stored them in MySQL
Used Python to extract database data, and completed data cleaning and column classification. Regular expressions to separate English, traditional Chinese, simplified Chinese and other different characters.
Used NLP training models, such as NLTK and Jieba for text training, through statistical analysis, increased the model words and paragraph cut accuracy. correction of model performed.
Counted the frequency of word occurrences, used Apriori to analyze the correlation between words and automatically exported the analysis results to an Excel report by openpyxl.
|
"The Cosmetic Trend From Online Forums ."
COVID-19 has affected our day-to-day life, Taiwan's government announced strict actions to reduce the transmission in communities. We know the cosmetic industries are affected by COVID-19 this year. I want to know the most popular products and brands.
I used correlation analysis to know the most frequently discussed brands and cosmetic products to understand users' needs. Then using natural language analysis to understand the audience's favorite brand and conduct long-term tracking.
I crawled the contents of cosmetic topic discussions for analysis from Taiwan's most popular public forums: PTT and Dcard.
Used beautiful soup by Python to extract the forum discussions, parse out the Post ID and connect to the API provided by the platform to capture the content of each post.
Used natural language processing to divide word structures and count the product category. Found association rules between products and brands.
From the analysis results, plot the number of PTT and Dcard posts on a linear graph, and added the trend line. Which shows the cosmetic discussion from PTT were gradually decreasing.
The core of recent discussions are lip gloss and base makeup. The result also showed the brand relevance in various products. The most popular lip gloss brand is Christian Dior.
As a speaker for Crowdsource by Google, he gave a talk on the topic of "Machine Learning" at the 2020 Developers Conference.
Google Developers Groups
The statistical and correlation analysis of the video game sales market reveals that the performance of action games has been growing in recent years.
Youtube
PHP with mySQL to build a membership database, providing users with the ability to register for an account while on the web, with account validation.
Youtube
Python for Yahoo weather web crawler to get date, temperature, rainfall chance for 10 days. Crawl Yahoo shopping sites with Python's Beautifulsoup to compile product search results and items for future price trend analysis.
Youtube
Python retrieves data from tables to obtain Goodinfo's daily stock market transactions and information about rallies and declines.
Youtube
Managed fans page on Facebook and used analytics to track posting performance and reply to audience feedback.
FacebookLooking for a job in Asia