# Configure Fuzzy Matching Threshold

When configuring user queries in the graph, we often encounter such embarrassing situations: the generalization ability is not strong through single sentence configurationBy asking the rules of the configuration, you are afraid that you cannot enumerate all the sentences. If the fuzzy matching can be carried out through the sentence pattern model, the above two problems will be solved.

It is now configured through the semantic similarity threshold in the robot settings in the weixin Dialogue Open Platform. Here the higher the similarity threshold, the harder it is to match, the lower the generalization ability is stronger (if the threshold is configured to 0, keyword matching can be achieved if the threshold is configured to (1) to be exact)

Examples:

The questions and answers are as follows:

• The semantic similarity threshold is configured to 0.95 time

• The semantic similarity threshold is configured to 0.7 time

In this way, we achieve the generalization ability of single sentence through the semantic similarity model, and do not need to configure tedious questions.

However, it is worth noting that the lower the semantic similarity threshold, the stronger the generalization ability, and there will be some risks, such as in multi-intention, there will be similar situations, when the threshold is adjusted very low, it will lead to the intention to hit disorder.