The hit-test result is not right #174

Closed
opened 2026-02-21 17:26:14 -05:00 by yindo · 4 comments
Owner

Originally created by @saifeiLee on GitHub (Jun 15, 2023).

I use the original question in my document to test. But Dify can not hit the corresponding text chunk. Here is the original text chunk in dataset:
image

Here is the hitting-test result:
image

Dify version: Cloud

Steps To Reproduce

  1. The original dataset can be downloaded here: Test Dataset
  2. Upload to dataset and test with keyword: "酒店找谁"

The current behavior

The hit result is not what we expected.

The expected behavior

Dify should hit the text chunk contains the keyword “酒店找谁”

Originally created by @saifeiLee on GitHub (Jun 15, 2023). I use the original question in my document to test. But Dify can not hit the corresponding text chunk. Here is the original text chunk in dataset: <img width="1069" alt="image" src="https://github.com/langgenius/dify/assets/28710808/88044f46-dd1b-4cc5-8094-0f50e825dcd8"> Here is the hitting-test result: <img width="1225" alt="image" src="https://github.com/langgenius/dify/assets/28710808/36309588-4ddf-4a9f-9c04-8986ba922934"> Dify version: Cloud ## Steps To Reproduce 1. The original dataset can be downloaded here: [Test Dataset](https://docs.google.com/spreadsheets/d/1Zy9Fag-DmadBeRU-Kur7KHRX4D3-gzSN/edit#gid=567952901) 2. Upload to dataset and test with keyword: "酒店找谁" ## The current behavior The hit result is not what we expected. ## The expected behavior Dify should hit the text chunk contains the keyword “酒店找谁”
yindo added the 🐞 bug label 2026-02-21 17:26:14 -05:00
yindo closed this issue 2026-02-21 17:26:14 -05:00
Author
Owner

@cmastro commented on GitHub (Jun 17, 2023):

I have the same issue. If I use some dataset, the answer is wrong or incomplete.

@cmastro commented on GitHub (Jun 17, 2023): I have the same issue. If I use some dataset, the answer is wrong or incomplete.
Author
Owner

@allentofight commented on GitHub (Jun 18, 2023):

+1, the same issue

@allentofight commented on GitHub (Jun 18, 2023): +1, the same issue
Author
Owner

@saifeiLee commented on GitHub (Jun 20, 2023):

I found that if we exclude redundant text in dataset. The query result can be more accurate.

@saifeiLee commented on GitHub (Jun 20, 2023): I found that if we exclude redundant text in dataset. The query result can be more accurate.
Author
Owner

@saifeiLee commented on GitHub (Jun 21, 2023):

Here is the result of the comparison test.

  1. With 'None':
    image

  2. Without 'None':
    image

@saifeiLee commented on GitHub (Jun 21, 2023): Here is the result of the comparison test. 1. With 'None': ![image](https://github.com/langgenius/dify/assets/28710808/4ab7a1f5-de24-4983-a642-84e495b88eaf) 2. Without 'None': ![image](https://github.com/langgenius/dify/assets/28710808/74ce52ef-9dbc-49e0-af70-4ff06acd7494)
Sign in to join this conversation.
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: langgenius/dify#174