[Bug]: cite not effect in agent generate component

[samuelyyl]

Self Checks

• I have searched for existing issues search for existing issues, including closed ones.
• I confirm that I am using English to submit this report (Language Policy).
• Non-english title submitions will be closed directly ( 非英文标题的提交将会被直接关闭 ) (Language Policy).
• Please do not modify this template :) and fill in all the required fields.

RAGFlow workspace code commit ID

de5727f

RAGFlow image version

build image with main branch lastest commit

Other environment information

not about env

Actual behavior

In the file agent/component/generate.py, the variables content_ltks and vector are never created or included in retrieval_res.

      if self._param.cite and "content_ltks" in retrieval_res.columns and "vector" in retrieval_res.columns:
            res = self.set_cite(retrieval_res, ans)
            return pd.DataFrame([res])

Expected behavior

No response

Steps to reproduce

create any agent using retrieval component and generate component can reproduce .

Additional information

No response

[dosubot]

Hello @samuelyyl! I'm Dosu, a bot here to help you solve bugs, answer questions, and assist you in becoming a contributor. I'm ready to support you while you wait for a human maintainer.

The issue you're encountering seems to be related to the cite functionality in the generate component of RAGFlow. The problem is that the content_ltks and vector columns are not being created or included in the retrieval_res DataFrame, which is necessary for the set_cite method to function properly.

To address this, you should ensure that the retrieval_res DataFrame is correctly populated with these columns before it reaches the generate component. This might involve checking the retrieval component's output to ensure it includes the necessary data or modifying the data transformation process to include these columns.

The set_cite method relies on these columns to insert citations into the generated answer. It processes the retrieval_res DataFrame by dropping rows with missing values in these columns and using them to add citations to the answer ^[1].

You may need to trace back to where the retrieval_res DataFrame is constructed and ensure that the content_ltks and vector columns are being populated with the expected data. If these columns are not being generated, you might need to adjust the retrieval logic or data processing steps to include them.

To continue talking to Dosu, mention @dosu.

---

^{Help Dosu learn! Give it feedback: Great Response | Irrelevant Answer | Incorrect Sources | Too Verbose | Hallucination | Bug Report | Other}

[samuelyyl]

It seems that the issue is caused by the agent/component/retrieval.py file. The following code:

        df = pd.DataFrame({"content": kb_prompt(kbinfos, 200000)})
        logging.debug("{} {}".format(query, df))
        return df.dropna()

The function kb_prompt converts kbinfos into a string list, which results in the loss of important information related to retrieval. This could be the reason why the cite functionality in the generate.py component is not triggered correctly.”

Additionally, the current agent/retrieval and generate components seem to behave similarly to the chat API but are not exactly the same. It might be worth considering adding a component that fully mimics the behavior of the chat API to address this inconsistency.

[dosubot]

The issue you're facing seems to stem from the way kb_prompt processes kbinfos in the agent/component/retrieval.py file. The kb_prompt function transforms kbinfos into a list of formatted strings, which may lead to the loss of detailed information necessary for citation functionality, such as content_ltks and vector data ^[1].

To resolve this, you might need to modify the retrieval process to ensure that content_ltks and vector are included in the retrieval_res DataFrame. This could involve adjusting how kbinfos is processed or ensuring that these columns are populated before the data is passed to the generate component.

Regarding the behavior of the agent components compared to the chat API, if the current components do not fully mimic the chat API, you might consider implementing a new component or modifying existing ones to align their behavior more closely with the chat API. This could help address inconsistencies and improve functionality. However, specific details on the differences between the components and the chat API are not available ^[1].

To continue talking to Dosu, mention @dosu.

---

^{Help Dosu learn! Give it feedback: Great Response | Irrelevant Answer | Incorrect Sources | Too Verbose | Hallucination | Bug Report | Other}