File size: 9,944 Bytes
e8b9514 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 |
import chromadb
from chromadb import Documents, EmbeddingFunction, Embeddings
from transformers import AutoModel
import json
from numpy.linalg import norm
import sqlite3
import urllib
# this module act as a singleton class
class JinaAIEmbeddingFunction(EmbeddingFunction):
def __init__(self, model):
super().__init__()
self.model = model
def __call__(self, input: Documents) -> Embeddings:
embeddings = self.model.encode(input)
return embeddings.tolist()
# instance of embedding_model
embedding_model = AutoModel.from_pretrained('jinaai/jina-embeddings-v2-base-en',
trust_remote_code=True,
cache_dir='models')
# instance of JinaAIEmbeddingFunction
ef = JinaAIEmbeddingFunction(embedding_model)
# list of topics
topic_descriptions = json.load(open("topic_descriptions.txt"))
topics = list(dict.keys(topic_descriptions))
embeddings = [embedding_model.encode(topic_descriptions[key]) for key in topic_descriptions]
cos_sim = lambda a,b: (a @ b.T) / (norm(a)*norm(b))
def choose_topic(summary):
embed = embedding_model.encode(summary)
topic = ""
max_sim = 0.
for i,key in enumerate(topics):
sim = cos_sim(embed,embeddings[i])
if sim > max_sim:
topic = key
max_sim = sim
return topic
def authors_list_to_str(authors):
"""input a list of authors, return a string represent authors"""
text = ""
for author in authors:
text+=author+", "
return text[:-3]
def authors_str_to_list(string):
"""input a string of authors, return a list of authors"""
authors = []
list_auth = string.split("and")
for author in list_auth:
if author != "et al.":
authors.append(author.strip())
return authors
def chunk_texts(text, max_char=400):
"""
Chunk a long text into several chunks, with each chunk about 300-400 characters long,
but make sure no word is cut in half.
Args:
text: The long text to be chunked.
max_char: The maximum number of characters per chunk (default: 400).
Returns:
A list of chunks.
"""
chunks = []
current_chunk = ""
words = text.split()
for word in words:
if len(current_chunk) + len(word) + 1 >= max_char:
chunks.append(current_chunk)
current_chunk = " "
else:
current_chunk += " " + word
chunks.append(current_chunk.strip())
return chunks
def trimming(txt):
start = txt.find("{")
end = txt.rfind("}")
return txt[start:end+1].replace("\n"," ")
# crawl data
def extract_tag(txt,tagname):
return txt[txt.find("<"+tagname+">")+len(tagname)+2:txt.find("</"+tagname+">")]
def get_record(extract):
id = extract_tag(extract,"id")
updated = extract_tag(extract,"updated")
published = extract_tag(extract,"published")
title = extract_tag(extract,"title").replace("\n ","").strip()
summary = extract_tag(extract,"summary").replace("\n","").strip()
authors = []
while extract.find("<author>")!=-1:
# author = extract[extract.find("<name>")+6:extract.find("</name>")]
author = extract_tag(extract,"name")
extract = extract[extract.find("</author>")+9:]
authors.append(author)
pattern = '<link title="pdf" href="'
link_start = extract.find('<link title="pdf" href="')
link = extract[link_start+len(pattern):extract.find("rel=",link_start)-2]
return [id, updated, published, title, authors, link, summary]
def crawl_exact_paper(title,author,max_results=3):
authors = authors_list_to_str(author)
records = []
url = 'http://export.arxiv.org/api/query?search_query=ti:{title}+AND+au:{author}&max_results={max_results}'.format(title=title,author=authors,max_results=max_results)
url = url.replace(" ","%20")
try:
arxiv_page = urllib.request.urlopen(url,timeout=100).read()
xml = str(arxiv_page,encoding="utf-8")
while xml.find("<entry>") != -1:
extract = xml[xml.find("<entry>")+7:xml.find("</entry>")]
xml = xml[xml.find("</entry>")+8:]
extract = get_record(extract)
topic = choose_topic(extract[6])
records.append([topic,*extract])
return records
except Exception as e:
return "Error: "+str(e)
def crawl_arxiv(keyword_list, max_results=100):
baseurl = 'http://export.arxiv.org/api/query?search_query='
records = []
for i,keyword in enumerate(keyword_list):
if i ==0:
url = baseurl + 'all:' + keyword
else:
url = url + '+OR+' + 'all:' + keyword
url = url+ '&max_results=' + str(max_results)
url = url.replace(' ', '%20')
try:
arxiv_page = urllib.request.urlopen(url,timeout=100).read()
xml = str(arxiv_page,encoding="utf-8")
while xml.find("<entry>") != -1:
extract = xml[xml.find("<entry>")+7:xml.find("</entry>")]
xml = xml[xml.find("</entry>")+8:]
extract = get_record(extract)
topic = choose_topic(extract[6])
records.append([topic,*extract])
return records
except Exception as e:
return "Error: "+str(e)
class ArxivSQL:
def __init__(self, table="arxivsql", name="arxiv_records_sql"):
self.con = sqlite3.connect(name)
self.cur = self.con.cursor()
self.table = table
def query(self, title="", author=[]):
if len(title)>0:
query_title = 'title like "%{}%"'.format(title)
else:
query_title = "True"
if len(author)>0:
query_author = 'authors like '
for auth in author:
query_author += "'%{}%' or ".format(auth)
query_author = query_author[:-4]
else:
query_author = "True"
query = "select * from {} where {} and {}".format(self.table,query_title,query_author)
result = self.cur.execute(query)
return result.fetchall()
def query_id(self, ids=[]):
try:
if len(ids) == 0:
return None
query = "select * from {} where id in (".format(self.table)
for id in ids:
query+="'"+id+"',"
query = query[:-1] + ")"
result = self.cur.execute(query)
return result.fetchall()
except Exception as e:
print(e)
print("Error query: ",query)
def add(self, crawl_records):
"""
Add crawl_records (list) obtained from arxiv_crawlers
A record is a list of 8 columns:
[topic, id, updated, published, title, author, link, summary]
Return the final length of the database table
"""
results = ""
for record in crawl_records:
try:
query = """insert into arxivsql values("{}","{}","{}","{}","{}","{}","{}")""".format(
record[1][21:],
record[0],
record[4].replace('"',"'"),
authors_list_to_str(record[5]),
record[2][:10],
record[3][:10],
record[6]
)
self.cur.execute(query)
self.con.commit()
except Exception as e:
result+=str(e)
result+="\n" + query + "\n"
finally:
return results
# instance of ArxivSQL
sqldb = ArxivSQL()
class ArxivChroma:
"""
Create an interface to arxivdb, which only support query and addition.
This interface do not support edition and deletion procedures.
"""
def __init__(self, table="arxiv_records", name="arxivdb/"):
self.client = chromadb.PersistentClient(name)
self.model = AutoModel.from_pretrained('jinaai/jina-embeddings-v2-base-en',
trust_remote_code=True,
cache_dir='models')
self.collection = self.client.get_or_create_collection(table,
embedding_function=JinaAIEmbeddingFunction(
model = self.model
))
def query_relevant(self, keywords, query_texts, n_results=3):
"""
Perform a query using a list of keywords (str),
or using a relavant string
"""
contains = []
for keyword in keywords:
contains.append({"$contains":keyword.lower()})
return self.collection.query(
query_texts=query_texts,
where_document={
"$or":contains
},
n_results=n_results,
)
def query_exact(self, id):
ids = ["{}_{}".format(id,j) for j in range(0,10)]
return self.collection.get(ids=ids)
def add(self, crawl_records):
"""
Add crawl_records (list) obtained from arxiv_crawlers
A record is a list of 8 columns:
[topic, id, updated, published, title, author, link, summary]
Return the final length of the database table
"""
for record in crawl_records:
embed_text = """
Topic: {},
Title: {},
Summary: {}
""".format(record[0],record[4],record[7])
chunks = chunk_texts(embed_text)
ids = [record[1][21:]+"_"+str(j) for j in range(len(chunks))]
paper_ids = [{"paper_id":record[1][21:]} for _ in range(len(chunks))]
self.collection.add(
documents = chunks,
metadatas=paper_ids,
ids = ids
)
return self.collection.count()
# instance of ArxivChroma
db = ArxivChroma()
|