Semantic Scholar Loader

bash

pip install llama-index-readers-semanticscholar

pip install llama-index-llms-openai

Welcome to Semantic Scholar Loader. This module serves as a crucial utility for researchers and professionals looking to get scholarly articles and publications from the Semantic Scholar database.

For any research topic you are interested in, this loader reads relevant papers from a search result in Semantic Scholar into Documents.

Please go through demo_s2.ipynb

Some preliminaries -

query_space : broad area of research
query_string : a specific question to the documents in the query space

UPDATE :

To download the open access pdfs and extract text from them, simply mark the full_text flag as True :

python

s2reader = SemanticScholarReader()
documents = s2reader.load_data(query_space, total_papers, full_text=True)

Usage

Here is an example of how to use this loader in llama_index and get citations for a given query.

LlamaIndex

python

from llama_index.llms.openai import OpenAI
from llama_index.core.query_engine import CitationQueryEngine
from llama_index.core import VectorStoreIndex, ServiceContext
from llama_index.readers.semanticscholar import SemanticScholarReader

s2reader = SemanticScholarReader()

# narrow down the search space
query_space = "large language models"

# increase limit to get more documents
documents = s2reader.load_data(query=query_space, limit=10)

service_context = ServiceContext.from_defaults(
    llm=OpenAI(model="gpt-3.5-turbo", temperature=0)
)
index = VectorStoreIndex.from_documents(
    documents, service_context=service_context
)

query_engine = CitationQueryEngine.from_args(
    index,
    similarity_top_k=3,
    citation_chunk_size=512,
)

# query the index
response = query_engine.query("limitations of using large language models")
print("Answer: ", response)
print("Source nodes: ")
for node in response.source_nodes:
    print(node.node.metadata)

Output

bash

Answer:  The limitations of using large language models include the struggle to learn long-tail knowledge [2], the need for scaling by many orders of magnitude to reach competitive performance on questions with little support in the pre-training data [2], and the difficulty in synthesizing complex programs from natural language descriptions [3].
Source nodes:
{'venue': 'arXiv.org', 'year': 2022, 'paperId': '3eed4de25636ac90f39f6e1ef70e3507ed61a2a6', 'citationCount': 35, 'openAccessPdf': None, 'authors': ['M. Shanahan'], 'title': 'Talking About Large Language Models'}
{'venue': 'arXiv.org', 'year': 2022, 'paperId': '6491980820d9c255b9d798874c8fce696750e0d9', 'citationCount': 31, 'openAccessPdf': None, 'authors': ['Nikhil Kandpal', 'H. Deng', 'Adam Roberts', 'Eric Wallace', 'Colin Raffel'], 'title': 'Large Language Models Struggle to Learn Long-Tail Knowledge'}
{'venue': 'arXiv.org', 'year': 2021, 'paperId': 'a38e0f993e4805ba8a9beae4c275c91ffcec01df', 'citationCount': 305, 'openAccessPdf': None, 'authors': ['Jacob Austin', 'Augustus Odena', 'Maxwell Nye', 'Maarten Bosma', 'H. Michalewski', 'David Dohan', 'Ellen Jiang', 'Carrie J. Cai', 'Michael Terry', 'Quoc V. Le', 'Charles Sutton'], 'title': 'Program Synthesis with Large Language Models'}