tatsu-lab/gpt_paper_assistant

GPT4 based personalized ArXiv paper assistant bot

gpt_paper_assistant's Issues

No new papers error

Hi, it seems like starting from today, the current implementation doesn't work anymore due to the following unexpected error:

No new papers since Fri, 17 Nov 2023 23:34:59 GMT for cs.CL
Traceback (most recent call last):
  File "/home/runner/work/GPT-Paper-Assistant/GPT-Paper-Assistant/main.py", line 193, in <module>
    papers = list(get_papers_from_arxiv(config))
  File "/home/runner/work/GPT-Paper-Assistant/GPT-Paper-Assistant/main.py", line 154, in get_papers_from_arxiv
    papers = get_papers_from_arxiv_rss_api(area.strip(), config)
  File "/home/runner/work/GPT-Paper-Assistant/GPT-Paper-Assistant/arxiv_scraper.py", line [12](https://github.com/sleepymalc/GPT-Paper-Assistant/actions/runs/6916931816/job/18817517300#step:5:13)6, in get_papers_from_arxiv_rss_api
    paper_list, timestamp, last_id = get_papers_from_arxiv_rss(area, config)
ValueError: not enough values to unpack (expected 3, got 0)
Error: Process completed with exit code 1.

Needless to say, after checking http://export.arxiv.org/rss/cs.CL, there does have some new papers today for cs.GL. Moreover, seems like your current deployment also encounters such problem.

ArXiv RSS vs. endpoint

Hi, thanks for the awesome project!

I noticed that both the ArXiv RSS feed and the endpoint are used to scrape ArXiv papers. The code mentions that the RSS feed is buggy, and sometimes drops new papers. Given this, wouldn't it be more reliable to exclusively use the endpoint instead?

Is GPT-4 a more effective recommender when compared to other methods for paper recommendation?

I'm intrigued by this system's use of GPT-4 score relevance for papers
Curious to hear your thoughts:
What other algorithms for paper recommendation are there, like arxiv-sanity?
Also, is GPT-4 stand a more effective recommender when compared to existing methodologies?

Support for GPT 3.5

Hi,

I do not have access to the Chat GPT4 premium plan but I generated an API key, and I encountered an error when running the workflow manually. How should I change config.ini to switch to the default Chat GPT 3.5 version?

Category classificaion and division

Now, all papers are listed together, which is hard to find when topics are too many. So I suggest adding classification and division to each topic.

Support Email

Can you add support for sending emails using github actions?

Local variable `cost` referenced before assignment

The most recent build (https://github.com/tatsu-lab/gpt_paper_assistant/actions/runs/7855086460) fails:

Number of papers:0
Getting author info for 0 authors
0it [00:00, ?it/s]
0it [00:00, ?it/s]
Traceback (most recent call last):
  File "/home/runner/work/gpt_paper_assistant/gpt_paper_assistant/main.py", line 220, in <module>
0 papers after hindex filtering
    filter_by_gpt(
  File "/home/runner/work/gpt_paper_assistant/gpt_paper_assistant/filter_papers.py", line 188, in filter_by_gpt
    paper_list, cost = filter_papers_by_title(
  File "/home/runner/work/gpt_paper_assistant/gpt_paper_assistant/filter_papers.py", line [14](https://github.com/tatsu-lab/gpt_paper_assistant/actions/runs/7855086460/job/21436449481#step:5:15)8, in filter_papers_by_title
    return final_list, cost
UnboundLocalError: local variable 'cost' referenced before assignment
Error: Process completed with exit code 1.

I also encountered the same bug.

IndexError: list index out of range

There's a IndexError: list index out of range in this run, and I also encountered it.
Another bug is that today's run shows that there's no new paper, which is again a bug I think.

Feature Request: Daily Archive of Paper Recommendations

Feature Description

Currently, it seems that the daily paper recommendations are only accessible through the GitHub page (if i understand it correctly), which limits the ability to revisit recommendations from previous days. This limitation makes it challenging to track and review papers that were recommended earlier.

Suggestion

I propose a feature request that automatically saves the daily recommendation results. This feature would provide users with the ability to access past recommendations at their convenience.

Issue with Irrelevant and Low Novelty Paper Recommendations

Description

Hello, I am highly interested in your project but have encountered some issues during execution. I forked the repository and ran the code manually without altering any configuration files, as per the instructions on GitHub. The execution resulted in six paper recommendations. However, upon reviewing the comments of these papers, it was evident that they were irrelevant to the given topic, scoring only 3-4 in both relevance and novelty.

Concern

This outcome is quite puzzling, especially since it contrasts with the high relevance seen in the examples provided in your repository. I am unsure about the cause of this discrepancy and would appreciate any guidance or insights you could offer.