Project background

  • The seller does not reply in time, which affects the transaction

  • The same problem sellers respond to uneven quality

  • The seller answers the frequently asked questions repeatedly

  • Sellers in the process of bargaining experience is poor

Technical framework

Property initializes the module


Quiz module
User question and answer knowledge extraction module


The core algorithm

Intention recognition

  • For the attributes that require high accuracy and do not allow wrong answers, we use the regular method as the front module.

  • For semantically diverse attributes, we use deep learning methods to identify as many user questions as possible.

Table1 comparison of main methods of intention recognition

Rules of pre –

Dependency classification model
BERT model

Words art generated

  • Identification of buyer’s query intention: For buyer’s query question, identify the corresponding query intention seller’s answer attribute extraction;

  • Through the sequence annotation method, the attribute value content is extracted from the seller’s answer.

  • Speech generation, according to the predefined template to fill in the attributes and attribute values, to generate a complete answer.


Application and Effect

Sample scenario

In conversation example 1, the buyer tried to communicate with the seller and negotiate the price. The seller replied to the user 12.5 hours after the buyer inquired, and the reply page was stiff, so the product did not close the deal at last.

Figure6 dialog example 1–The chatbot is not configured

Figure7 dialog example 2- start the chatbot

Business impact

  • Prompt response rate increase: prompt response rate increase by 30% within 2 hours.

  • More chat rounds: The number of sales review rounds increased by nearly 20%.

  • Lower seller response costs: Thousands of hours saved per 20 seconds.

  • Increased funnel efficiency: the funnel of commodities at all levels from exposure to interaction to transaction has increased, which promotes the circulation efficiency of commodities, and also verifies that improving chat efficiency in the community can effectively improve the conversion rate at all levels.

  • Improved sales rate of goods: compared with goods without chatbot enabled, the sales of goods with chatbot enabled increased by 30% in 7 days on average.

  • Improved interactive transaction conversion rate: Compared with the transaction conversion when the seller actually replies to the buyer, the conversion rate of the chatbot is higher than that when the seller replies more than 2 hours or more, the average conversion rate is increased by 30%.

The Idle Fish Team is the industry leader in the new technology of Flutter+Dart FaaS integration, right now! Client side/server side Java/architecture/front-end/quality engineer all look forward to your joining. Base Hangzhou Alibaba Xixi Park, we will work together to create community products with creative space, do top-level open source projects in depth, expand technological boundaries and achieve the ultimate!

* Send resumes to small idle fish → [email protected]


More series of articles, open source projects, key insights, in-depth interpretation

Please look for the idle fish technology