Responsibilities
Team introduction: ByteDance's Douyin search team is mainly responsible for Douyin search algorithm innovation and architecture research and development, mainly including short video, live broadcast, local life, visual search and other business lines. We use the most cutting-edge machine learning technology for end-to-end modeling and continue to innovate and make breakthroughs. At the same time, we focus on the construction and performance optimization of distributed systems and machine learning systems, from memory and disk optimization to the exploration of index compression, recall, sorting and other algorithms, fully providing students with opportunities to grow themselves. The main work directions include: 1. Exploring cutting-edge NLP technology: from basic word segmentation, NER, to application Query analysis, basic correlation, etc., applying deep learning models in the entire link, every detail is full of challenges 2. Cross-modal matching technology: applying CV+NLP deep learning technology in search to give video search a more powerful retrieval capability 3. Large-scale streaming machine learning technology: Apply large-scale machine learning to solve the recommendation problem in search, making the search more personalized and understanding of you 4. Architecture with hundreds of billions of data scale: In-depth research and innovation in all aspects from large-scale offline computing, performance and scheduling optimization of distributed systems to building high-availability, high-throughput and low-latency online services. 1. Participate in the improvement of search recommendation models and strategies for Douyin, Toutiao, Xigua Video, Jianying and other apps, as well as domestic e-commerce, life services and other key businesses, and be responsible for the search traffic and user penetration growth of these businesses & search mind building tasks 2. With recommendation algorithms as the core technology stack, improve the recommendation system based on ultra-large-scale machine learning models, covering the full link technology links from candidate mining to recall, rough ranking, fine ranking, multi-objective fusion 3. Explore the upper limit of short text recommendation and general recommendation technology, focusing on the joint application of recommendation and NLP technology, as well as the exploration of cutting-edge technologies such as multi-modality.
Qualifications
1. Basic requirements: good algorithm design capabilities and engineering implementation capabilities, and practical experience in machine learning/reinforcement learning/NLP applications (Part 1) 2. Have good communication and expression skills, and have a clear understanding of good user experience. Candidates with good product awareness are preferred. Bonus points: 1. Applicants with experience in optimizing large-scale (large candidates, large number of users) recommendation systems are preferred 2. Applicants with experience in pre-training large-scale NLP language models are preferred 3. Applicants who have participated in competitions such as ACM and achieved good rankings are preferred.
