Baidu Kaoba: A Case Study of Natural Language Processing and Semantic Analysis

On June 28, Dr. Zhao Shiqi, senior researcher at Baidu’s Natural Language Processing (NLP) department, delivered a talk at the 51st Baidu Technology Salon, focusing on the current state of NLP and semantic analysis technologies. He used Baidu’s lightweight application “Baidu Kaoba”—a conversational search tool for college admission information—as a concrete example.

Zhao highlighted the growing prevalence of NLP‑driven products such as Siri, chatbots, and Xiaoice, emphasizing that future internet services will increasingly rely on semantic analysis. He described Baidu Kaoba as integrating several core NLP techniques: inference‑based rewriting, anaphora resolution with ellipsis completion, concept clarification, and interactive generation.

Inference‑based rewriting transforms casual spoken queries into database‑friendly statements (e.g., converting “What university can I get into with a score of 630?” into “Universities with a score line below 630”).

Ellipsis completion supplies missing entities for knowledge‑base queries, offering low‑cost resource allocation and strong domain transferability.

Concept clarification resolves ambiguous references, such as distinguishing whether “Shan Da” refers to Shanxi University or Shandong University.

Zhao stressed that Baidu aims beyond shallow interactions; the goal is deep, domain‑specific understanding that can answer any user question within a chosen field. He likened current conversational search to the “Bumble” robot in *Iron Man*—still imperfect but improving through continuous interaction.

The talk concluded with audience feedback praising the depth of Baidu’s NLP integration and expressing strong interest in further technical details. Baidu’s Technology Salon, now in its fourth year and 51st session, continues to promote open technical exchange among China’s high‑end internet professionals.