initial smoke test

.gitignore

@@ -0,0 +1 @@
+.streamlit

README.md

@@ -1,5 +1,5 @@
 ---
-title: Scite Qa Demo
+title: CitationQA - Answers from science with citations
 emoji: 🐠
 colorFrom: pink
 colorTo: indigo

app.py

@@ -0,0 +1,128 @@
+import streamlit as st
+from transformers import pipeline
+import requests
+from bs4 import BeautifulSoup
+
+SCITE_API_KEY = st.secrets["SCITE_API_KEY"]
+
+def remove_html(x):
+    soup = BeautifulSoup(x, 'html.parser')
+    text = soup.get_text()
+    return text
+
+def search(term, limit=10):
+    search = f"https://api.scite.ai/search?mode=citations&term={term}&limit={limit}&offset=0&user_slug=domenic-rosati-keW5&compute_aggregations=false"
+    req = requests.get(
+        search,
+        headers={
+            'Authorization': f'Bearer {SCITE_API_KEY}'
+        }
+    )
+    return (
+        remove_html('\n'.join(['\n'.join([cite['snippet'] for cite in doc['citations']]) for doc in req.json()['hits']])),
+        [(doc['doi'], doc['citations'], doc['title']) for doc in req.json()['hits']]
+    )
+
+
+def find_source(text, docs):
+    for doc in docs:
+        if text in remove_html(doc[1][0]['snippet']):
+            new_text = text
+            for snip in remove_html(doc[1][0]['snippet']).split('.'):
+                if text in snip:
+                    new_text = snip
+            return {
+                'citation_statement': doc[1][0]['snippet'].replace('<strong class="highlight">', '').replace('</strong>', ''),
+                'text': new_text,
+                'from': doc[1][0]['source'],
+                'supporting': doc[1][0]['target'],
+                'source_title': doc[2],
+                'source_link': f"https://scite.ai/reports/{doc[0]}"
+            }
+    return {
+            'citation_statement': '',
+            'text': text,
+            'from': '',
+            'supporting': ''
+        }
+
+@st.experimental_singleton
+def init_models():
+    question_answerer = pipeline("question-answering", model='sultan/BioM-ELECTRA-Large-SQuAD2-BioASQ8B')
+    return question_answerer
+
+qa_model = init_models()
+
+
+def card(title, context, score, link):
+    return st.markdown(f"""
+    <div class="container-fluid">
+        <div class="row align-items-start">
+             <div  class="col-md-12 col-sm-12">
+                 <br>
+                 <span>
+                     {context}
+                     [<b>Score: </b>{score}]
+                 </span>
+                 <br>
+                 <b>From <a href="{link}">{title}</a></b>
+             </div>
+        </div>
+     </div>
+        """, unsafe_allow_html=True)
+
+st.title("Scientific Question Answering with Citations")
+
+st.write("""
+Ask a scientific question and get an answer drawn from [scite.ai](https://scite.ai) corpus of over 1.1bn citation statements.
+Answers are linked to source documents containing citations where users can explore further evidence from scientific literature for the answer.
+""")
+
+st.markdown("""
+<link rel="stylesheet" href="https://cdn.jsdelivr.net/npm/bootstrap@4.0.0/dist/css/bootstrap.min.css" integrity="sha384-Gn5384xqQ1aoWXA+058RXPxPg6fy4IWvTNh0E263XmFcJlSAwiGgFAW/dAiS6JXm" crossorigin="anonymous">
+""", unsafe_allow_html=True)
+
+def run_query(query):
+    context, orig_docs = search(query)
+    if not context.strip():
+        return st.markdown("""
+        <div class="container-fluid">
+        <div class="row align-items-start">
+             <div  class="col-md-12 col-sm-12">
+            Sorry... no results for that question! Try another.
+         </div>
+        </div>
+     </div>
+        """)
+
+    results = []
+    model_results = qa_model(question=query, context=context, top_k=10)
+    for result in model_results:
+        support = find_source(result['answer'], orig_docs)
+        results.append({
+            "answer": support['text'],
+            "title": support['source_title'],
+            "link": support['source_link'],
+            "context": support['citation_statement'],
+            "score": result['score']
+        })
+
+
+
+    sorted_result = sorted(results, key=lambda x: x['score'], reverse=True)
+    sorted_result = list({
+        result['context']: result for result in sorted_result
+    }.values())
+
+
+    for r in sorted_result:
+        answer = r["answer"]
+        ctx = r["context"].replace(answer, f"<mark>{answer}</mark>").replace('<cite', '<a').replace('</cite', '</a').replace('data-doi="', 'href="https://scite.ai/reports/')
+        title = r["title"].replace("_", " ")
+        score = round(r["score"], 4)
+        card(title, ctx, score, r['link'])
+
+query = st.text_input("Ask scientific literature a question", "")
+
+if query != "":
+    run_query(query)

requirements.txt

@@ -0,0 +1,4 @@
+transformers
+requests
+beautifulsoup4
+streamlit==1.2.0