forked from Ponysearch/Ponysearch
137 lines
4.8 KiB
Python
137 lines
4.8 KiB
Python
import json
|
|
from urllib import urlencode
|
|
from lxml import html
|
|
from searx.engines.xpath import extract_text
|
|
|
|
url = 'https://api.duckduckgo.com/?{query}&format=json&pretty=0&no_redirect=1&d=1'
|
|
|
|
def result_to_text(url, text, htmlResult):
|
|
# TODO : remove result ending with "Meaning" or "Category"
|
|
dom = html.fromstring(htmlResult)
|
|
a = dom.xpath('//a')
|
|
if len(a)>=1:
|
|
return extract_text(a[0])
|
|
else:
|
|
return text
|
|
|
|
def html_to_text(htmlFragment):
|
|
dom = html.fromstring(htmlFragment)
|
|
return extract_text(dom)
|
|
|
|
def request(query, params):
|
|
# TODO add kl={locale}
|
|
params['url'] = url.format(query=urlencode({'q': query}))
|
|
return params
|
|
|
|
|
|
def response(resp):
|
|
search_res = json.loads(resp.text)
|
|
results = []
|
|
|
|
content = ''
|
|
heading = search_res.get('Heading', '')
|
|
attributes = []
|
|
urls = []
|
|
infobox_id = None
|
|
relatedTopics = []
|
|
|
|
# add answer if there is one
|
|
answer = search_res.get('Answer', '')
|
|
if answer != '':
|
|
results.append({ 'answer' : html_to_text(answer) })
|
|
|
|
# add infobox
|
|
if 'Definition' in search_res:
|
|
content = content + search_res.get('Definition', '')
|
|
|
|
if 'Abstract' in search_res:
|
|
content = content + search_res.get('Abstract', '')
|
|
|
|
|
|
# image
|
|
image = search_res.get('Image', '')
|
|
image = None if image == '' else image
|
|
|
|
# attributes
|
|
if 'Infobox' in search_res:
|
|
infobox = search_res.get('Infobox', None)
|
|
if 'content' in infobox:
|
|
for info in infobox.get('content'):
|
|
attributes.append({'label': info.get('label'), 'value': info.get('value')})
|
|
|
|
# urls
|
|
for ddg_result in search_res.get('Results', []):
|
|
if 'FirstURL' in ddg_result:
|
|
firstURL = ddg_result.get('FirstURL', '')
|
|
text = ddg_result.get('Text', '')
|
|
urls.append({'title':text, 'url':firstURL})
|
|
results.append({'title':heading, 'url': firstURL})
|
|
|
|
# related topics
|
|
for ddg_result in search_res.get('RelatedTopics', None):
|
|
if 'FirstURL' in ddg_result:
|
|
suggestion = result_to_text(ddg_result.get('FirstURL', None), ddg_result.get('Text', None), ddg_result.get('Result', None))
|
|
if suggestion != heading:
|
|
results.append({'suggestion': suggestion})
|
|
elif 'Topics' in ddg_result:
|
|
suggestions = []
|
|
relatedTopics.append({ 'name' : ddg_result.get('Name', ''), 'suggestions': suggestions })
|
|
for topic_result in ddg_result.get('Topics', []):
|
|
suggestion = result_to_text(topic_result.get('FirstURL', None), topic_result.get('Text', None), topic_result.get('Result', None))
|
|
if suggestion != heading:
|
|
suggestions.append(suggestion)
|
|
|
|
# abstract
|
|
abstractURL = search_res.get('AbstractURL', '')
|
|
if abstractURL != '':
|
|
# add as result ? problem always in english
|
|
infobox_id = abstractURL
|
|
urls.append({'title': search_res.get('AbstractSource'), 'url': abstractURL})
|
|
|
|
# definition
|
|
definitionURL = search_res.get('DefinitionURL', '')
|
|
if definitionURL != '':
|
|
# add as result ? as answer ? problem always in english
|
|
infobox_id = definitionURL
|
|
urls.append({'title': search_res.get('DefinitionSource'), 'url': definitionURL})
|
|
|
|
# entity
|
|
entity = search_res.get('Entity', None)
|
|
# TODO continent / country / department / location / waterfall / mountain range : link to map search, get weather, near by locations
|
|
# TODO musician : link to music search
|
|
# TODO concert tour : ??
|
|
# TODO film / actor / television / media franchise : links to IMDB / rottentomatoes (or scrap result)
|
|
# TODO music : link tu musicbrainz / last.fm
|
|
# TODO book : ??
|
|
# TODO artist / playwright : ??
|
|
# TODO compagny : ??
|
|
# TODO software / os : ??
|
|
# TODO software engineer : ??
|
|
# TODO prepared food : ??
|
|
# TODO website : ??
|
|
# TODO performing art : ??
|
|
# TODO prepared food : ??
|
|
# TODO programming language : ??
|
|
# TODO file format : ??
|
|
|
|
if len(heading)>0:
|
|
# TODO get infobox.meta.value where .label='article_title'
|
|
if image==None and len(attributes)==0 and len(urls)==1 and len(relatedTopics)==0 and len(content)==0:
|
|
results.append({
|
|
'url': urls[0]['url'],
|
|
'title': heading,
|
|
'content': content
|
|
})
|
|
else:
|
|
results.append({
|
|
'infobox': heading,
|
|
'id': infobox_id,
|
|
'entity': entity,
|
|
'content': content,
|
|
'img_src' : image,
|
|
'attributes': attributes,
|
|
'urls': urls,
|
|
'relatedTopics': relatedTopics
|
|
})
|
|
|
|
return results
|