Interview with Li Benyang: AI, Knowledge Graphs, and Career Insights in Intelligent Recommendation

"Technology People Interview" is a 51CTO series that features in‑depth conversations with industry experts, revealing their experiences, values, and professional insights.

Li Benyang, a Harbin Institute of Technology alumnus specializing in natural language processing, has built a career around AI, working at Baidu, a big‑data startup, a medical‑tech firm, and now leading Autohome's intelligent recommendation content understanding team.

He discusses the rise of intelligent recommendation in the era of massive data, noting challenges such as cold start, accuracy, diversity, and the rapid emergence of multi‑task learning, multimodal fusion, and graph‑based feature learning.

Content understanding is essential for recommendation; Li outlines three main directions his team pursues: content profiling (tags, quality, similarity, sentiment), content ecosystem analysis (linking production and consumption), and knowledge‑graph construction centered on automobiles.

Content profiling includes tag systems, classification, concept tags, interest‑word tags, quality assessment, low‑quality detection, and similarity calculations for text and images.

Content ecosystem analysis uses multidimensional tags to improve user‑liked content production.

Automotive‑centric knowledge graphs support various downstream applications.

He explains how knowledge graphs enrich item and user representations, enabling deeper exploration beyond surface recommendations, such as linking a user's interest in a specific car model to broader preferences.

Li reflects on career decisions, emphasizing the benefits of large‑company experience for building professional habits, while also valuing startup experiences for rapid growth and broader perspective.

His advice for newcomers includes maintaining a learning mindset, avoiding frequent industry jumps, leveraging transferable technical skills, and deeply understanding industry context through communication with seasoned professionals.

Regarding management, Li contrasts the focus of engineers (technical solutions and output) with that of managers (team goals, coordination, and talent development), highlighting the importance of shared objectives, clear division of labor, and empathetic communication.

He shares his hiring criteria: personality (reliability), professional ability, and growth potential, with emphasis on fundamentals for recent graduates.

Looking ahead, Li predicts recommendation systems will become more multi‑modal, context‑aware, and regulated, with increased transparency and explainability to mitigate algorithmic black‑box concerns.