Design and Impact of a Chat Assistant for E‑commerce Negotiation and Greeting

Effective communication can significantly increase transaction probability in second‑hand marketplaces. Analysis of Xianyu chat logs shows that sellers who engage more frequently achieve higher conversion rates, while individual sellers often suffer from low conversion due to stiff or unemotional replies.

Price negotiation occupies 30%‑40% of chat content and usually appears at the later stage of a transaction. Improving the negotiation dialogue therefore offers a direct path to boost successful deals.

The solution adopts an asynchronous MQ consumption pipeline to avoid affecting the main messaging flow. Preliminary filters (message type, sender, keywords, fatigue control) reduce algorithmic load. Integrated price‑strength data provides reference and suggested prices, enabling the system to generate friendly bargaining scripts or reject unreasonable offers.

A generic intent‑recognition framework was abstracted using an IntentProcessor interface. Each intent implements filterAndCompleteContext for pre‑screening and process for business logic. The framework currently supports bargaining, greeting, and Q&A intents, selecting the most relevant intent by priority when multiple matches occur.

Experimental results show a 4% increase in reply rate and a 4% uplift in conversion rate for the bargaining scenario compared with a control group, confirming the hypothesis that guided conversation promotes transactions.

Future work will extend the assistant to additional intents such as free‑shipping and shipping‑origin detection, and deepen product‑information extraction to further assist sellers in completing item details.