Yiyan J.

Key Skills

Software

Other

Top Subject Matter

Vehicle related

Service related

Internet related

Top Data Types

Text

Top Task Types

Question Answering

Evaluation/Rating

Freelancer Overview

For data labeling: I previously worked at Microsoft on news headline and category annotation for Bing News across seven markets, including the United States, Germany, France, Australia, China, Japan, and South Korea. We began by conducting a competitive analysis of headlines and category news, followed by definition and categorization. Finally, we cleaned the data from news articles collected by Bing and distributed them in batches. We collaborated with freelancers to complete the multidimensional scoring and annotation of all news articles. For AI training data:I previously worked on AI data training at my first company, which specialized in forecasting future energy consumption based on actual electricity usage data from major commercial real estate groups in China, as well as other variables that might influence energy consumption. The input data included hourly-level metrics from the past decade for a total of 500 large commercial buildings operated by companies such as Wanda Group, China Resources Group, China Overseas Land & Investment, and Longfor Group. This data covered electricity consumption by appliance category, foot traffic, local temperatures, and holiday cycles. Ultimately, we used a neural network to develop an annual electricity consumption prediction model for large shopping malls.

Labeling Experience

Bing News Headline and Category News Quality Rating

TextEvaluation/Rating

1. Project Scope Responsible for manual quality control of Microsoft Bing’s top news stories and news across various category channels, including screening high-quality news, filtering out non-compliant and low-quality content, and producing annotated data to optimize news recommendations. 2. Specific Annotation and Quality Control Tasks Top News: Verify the timeliness, authority, and risk factors of candidate news stories, and assign news value ratings; Category News: Assign news articles to the appropriate channel categories, distinguish high-quality news, and flag non-compliant content such as advertisements and misinformation; 3. Project Scale A team of 10 members and 100 freelancers, divided into Top News, Category, and Review groups; average daily review of 20,000 news articles; 19 months of project participation, with a cumulative total of nearly 8.5 million annotated samples. 4. Quality Control Standards Personnel: Pre-employment training + practical assessment required for qualification; new hires undergo a 15-day full-inspection probationary period; Review: 15% of general news articles are randomly selected for re-inspection; 100% of headline news articles undergo re-review; annotation discrepancies are arbitrated by designated personnel; Metrics: General news annotation accuracy ≥98%; headlines ≥99.2%; missed detection rate for non-compliant content below 0.3%; daily review and rectification of mislabeled items.

2020 - 2021

Education

Work History