FB Reels Recommendation

FB Reels Recommendation

1. Tech lead of the recommendation department of short form video(The product is like TikTok), specialized in ranking algorithm, also served as the team manager of the ranking team. The DAU of the product grew from 3 million DAU to 30 million DAU, with 100%+ increase in watch time per user.

2. Tech lead of the search department of Tencent Video(The product is like Netflix+Youtube), leading all the applied algorithms in video search, including search ranking, retrieval, query understanding, and knowledge graph. As a result of algorithm improvement, the watch time per user is increased by 25% cumulatively.

Only went through the bootcamp, had to quit and returned to Beijing China because of an urgent health issue of my grandpa.

Search Ranking and Guest Targeting

• Launched Plus booker model for guest targeting, it increased Plus bookings globally by 8.7% and Plus booking value by 10.0%

• Built a search reranker to promote Plus bookings.

NLP on guest review and message between hosts and guests

• Launched a XLNet based model for aspects extraction, it is used to detect how issues happened during a trip.

• Launched a BERT based model for sentiment analysis, it is used to highlight potentially negative reviews and guest messages for human quality assessors to decide remedy actions

Computer Vision model for Plus listing selections

• Built a series of ResNet based CV model to select high quality listings based of listing photos. There is one model for each room type. The output of the models are then aggregated to decide if a listing should be invited to Plus.