| 
									
										
										
										
											2021-03-01 15:02:11 +01:00
										 |  |  | # SPDX-License-Identifier: AGPL-3.0-or-later | 
					
						
							| 
									
										
										
										
											2021-04-26 20:18:20 +02:00
										 |  |  | # lint: pylint | 
					
						
							| 
									
										
										
										
											2022-12-04 22:57:22 +01:00
										 |  |  | """This is the implementation of the Google Scholar engine.
 | 
					
						
							| 
									
										
										
										
											2021-03-01 15:02:11 +01:00
										 |  |  | 
 | 
					
						
							| 
									
										
										
										
											2022-12-04 22:57:22 +01:00
										 |  |  | Compared to other Google services the Scholar engine has a simple GET REST-API | 
					
						
							|  |  |  | and there does not exists `async` API.  Even though the API slightly vintage we | 
					
						
							|  |  |  | can make use of the :ref:`google API` to assemble the arguments of the GET | 
					
						
							|  |  |  | request. | 
					
						
							| 
									
										
										
										
											2021-03-01 15:02:11 +01:00
										 |  |  | """
 | 
					
						
							|  |  |  | 
 | 
					
						
							| 
									
										
										
										
											2022-12-04 22:57:22 +01:00
										 |  |  | from typing import TYPE_CHECKING | 
					
						
							|  |  |  | from typing import Optional | 
					
						
							| 
									
										
										
										
											2021-03-01 15:02:11 +01:00
										 |  |  | 
 | 
					
						
							|  |  |  | from urllib.parse import urlencode | 
					
						
							|  |  |  | from datetime import datetime | 
					
						
							|  |  |  | from lxml import html | 
					
						
							|  |  |  | 
 | 
					
						
							|  |  |  | from searx.utils import ( | 
					
						
							|  |  |  |     eval_xpath, | 
					
						
							| 
									
										
										
										
											2022-08-26 18:10:12 +02:00
										 |  |  |     eval_xpath_getindex, | 
					
						
							| 
									
										
										
										
											2021-03-01 15:02:11 +01:00
										 |  |  |     eval_xpath_list, | 
					
						
							|  |  |  |     extract_text, | 
					
						
							|  |  |  | ) | 
					
						
							|  |  |  | 
 | 
					
						
							| 
									
										
										
										
											2022-12-04 22:57:22 +01:00
										 |  |  | from searx.exceptions import SearxEngineCaptchaException | 
					
						
							|  |  |  | 
 | 
					
						
							|  |  |  | from searx.engines.google import fetch_traits  # pylint: disable=unused-import | 
					
						
							| 
									
										
										
										
											2021-03-01 15:02:11 +01:00
										 |  |  | from searx.engines.google import ( | 
					
						
							| 
									
										
										
										
											2022-12-04 22:57:22 +01:00
										 |  |  |     get_google_info, | 
					
						
							| 
									
										
										
										
											2021-03-01 15:02:11 +01:00
										 |  |  |     time_range_dict, | 
					
						
							|  |  |  | ) | 
					
						
							| 
									
										
										
										
											2022-12-04 22:57:22 +01:00
										 |  |  | from searx.enginelib.traits import EngineTraits | 
					
						
							| 
									
										
										
										
											2021-03-01 15:02:11 +01:00
										 |  |  | 
 | 
					
						
							| 
									
										
										
										
											2022-12-04 22:57:22 +01:00
										 |  |  | if TYPE_CHECKING: | 
					
						
							|  |  |  |     import logging | 
					
						
							| 
									
										
										
										
											2021-12-27 09:26:22 +01:00
										 |  |  | 
 | 
					
						
							| 
									
										
										
										
											2022-12-04 22:57:22 +01:00
										 |  |  |     logger: logging.Logger | 
					
						
							|  |  |  | 
 | 
					
						
							|  |  |  | traits: EngineTraits | 
					
						
							| 
									
										
										
										
											2021-03-01 15:02:11 +01:00
										 |  |  | 
 | 
					
						
							|  |  |  | # about | 
					
						
							|  |  |  | about = { | 
					
						
							|  |  |  |     "website": 'https://scholar.google.com', | 
					
						
							|  |  |  |     "wikidata_id": 'Q494817', | 
					
						
							|  |  |  |     "official_api_documentation": 'https://developers.google.com/custom-search', | 
					
						
							|  |  |  |     "use_official_api": False, | 
					
						
							|  |  |  |     "require_api_key": False, | 
					
						
							|  |  |  |     "results": 'HTML', | 
					
						
							|  |  |  | } | 
					
						
							|  |  |  | 
 | 
					
						
							|  |  |  | # engine dependent config | 
					
						
							| 
									
										
										
										
											2022-08-26 18:10:12 +02:00
										 |  |  | categories = ['science', 'scientific publications'] | 
					
						
							| 
									
										
										
										
											2021-03-01 15:02:11 +01:00
										 |  |  | paging = True | 
					
						
							|  |  |  | language_support = True | 
					
						
							|  |  |  | time_range_support = True | 
					
						
							|  |  |  | safesearch = False | 
					
						
							| 
									
										
										
										
											2022-08-01 17:01:59 +02:00
										 |  |  | send_accept_language_header = True | 
					
						
							| 
									
										
										
										
											2021-03-01 15:02:11 +01:00
										 |  |  | 
 | 
					
						
							| 
									
										
										
										
											2021-12-27 09:26:22 +01:00
										 |  |  | 
 | 
					
						
							| 
									
										
										
										
											2022-12-04 22:57:22 +01:00
										 |  |  | def time_range_args(params): | 
					
						
							|  |  |  |     """Returns a dictionary with a time range arguments based on
 | 
					
						
							|  |  |  |     ``params['time_range']``. | 
					
						
							| 
									
										
										
										
											2021-03-01 15:02:11 +01:00
										 |  |  | 
 | 
					
						
							| 
									
										
										
										
											2022-12-04 22:57:22 +01:00
										 |  |  |     Google Scholar supports a detailed search by year.  Searching by *last | 
					
						
							|  |  |  |     month* or *last week* (as offered by SearXNG) is uncommon for scientific | 
					
						
							|  |  |  |     publications and is not supported by Google Scholar. | 
					
						
							| 
									
										
										
										
											2021-03-01 15:02:11 +01:00
										 |  |  | 
 | 
					
						
							| 
									
										
										
										
											2022-12-04 22:57:22 +01:00
										 |  |  |     To limit the result list when the users selects a range, all the SearXNG | 
					
						
							|  |  |  |     ranges (*day*, *week*, *month*, *year*) are mapped to *year*.  If no range | 
					
						
							|  |  |  |     is set an empty dictionary of arguments is returned.  Example;  when | 
					
						
							|  |  |  |     user selects a time range (current year minus one in 2022): | 
					
						
							| 
									
										
										
										
											2021-03-01 15:02:11 +01:00
										 |  |  | 
 | 
					
						
							| 
									
										
										
										
											2022-12-04 22:57:22 +01:00
										 |  |  |     .. code:: python | 
					
						
							| 
									
										
										
										
											2021-03-01 15:02:11 +01:00
										 |  |  | 
 | 
					
						
							| 
									
										
										
										
											2022-12-04 22:57:22 +01:00
										 |  |  |         { 'as_ylo' : 2021 } | 
					
						
							| 
									
										
										
										
											2021-06-11 16:31:50 +02:00
										 |  |  | 
 | 
					
						
							| 
									
										
										
										
											2022-12-04 22:57:22 +01:00
										 |  |  |     """
 | 
					
						
							|  |  |  |     ret_val = {} | 
					
						
							|  |  |  |     if params['time_range'] in time_range_dict: | 
					
						
							|  |  |  |         ret_val['as_ylo'] = datetime.now().year - 1 | 
					
						
							|  |  |  |     return ret_val | 
					
						
							| 
									
										
										
										
											2021-03-01 15:02:11 +01:00
										 |  |  | 
 | 
					
						
							|  |  |  | 
 | 
					
						
							| 
									
										
										
										
											2022-12-04 22:57:22 +01:00
										 |  |  | def detect_google_captcha(dom): | 
					
						
							|  |  |  |     """In case of CAPTCHA Google Scholar open its own *not a Robot* dialog and is
 | 
					
						
							|  |  |  |     not redirected to ``sorry.google.com``. | 
					
						
							|  |  |  |     """
 | 
					
						
							|  |  |  |     if eval_xpath(dom, "//form[@id='gs_captcha_f']"): | 
					
						
							|  |  |  |         raise SearxEngineCaptchaException() | 
					
						
							|  |  |  | 
 | 
					
						
							| 
									
										
										
										
											2021-03-01 15:02:11 +01:00
										 |  |  | 
 | 
					
						
							| 
									
										
										
										
											2022-12-04 22:57:22 +01:00
										 |  |  | def request(query, params): | 
					
						
							|  |  |  |     """Google-Scholar search request""" | 
					
						
							| 
									
										
										
										
											2021-03-01 15:02:11 +01:00
										 |  |  | 
 | 
					
						
							| 
									
										
										
										
											2022-12-04 22:57:22 +01:00
										 |  |  |     google_info = get_google_info(params, traits) | 
					
						
							|  |  |  |     # subdomain is: scholar.google.xy | 
					
						
							|  |  |  |     google_info['subdomain'] = google_info['subdomain'].replace("www.", "scholar.") | 
					
						
							|  |  |  | 
 | 
					
						
							|  |  |  |     args = { | 
					
						
							|  |  |  |         'q': query, | 
					
						
							|  |  |  |         **google_info['params'], | 
					
						
							|  |  |  |         'start': (params['pageno'] - 1) * 10, | 
					
						
							|  |  |  |         'as_sdt': '2007',  # include patents / to disable set '0,5' | 
					
						
							|  |  |  |         'as_vis': '0',  # include citations / to disable set '1' | 
					
						
							|  |  |  |     } | 
					
						
							|  |  |  |     args.update(time_range_args(params)) | 
					
						
							|  |  |  | 
 | 
					
						
							|  |  |  |     params['url'] = 'https://' + google_info['subdomain'] + '/scholar?' + urlencode(args) | 
					
						
							|  |  |  |     params['cookies'] = google_info['cookies'] | 
					
						
							|  |  |  |     params['headers'].update(google_info['headers']) | 
					
						
							| 
									
										
										
										
											2021-03-01 15:02:11 +01:00
										 |  |  |     return params | 
					
						
							|  |  |  | 
 | 
					
						
							| 
									
										
										
										
											2021-12-27 09:26:22 +01:00
										 |  |  | 
 | 
					
						
							| 
									
										
										
										
											2022-08-26 18:10:12 +02:00
										 |  |  | def parse_gs_a(text: Optional[str]): | 
					
						
							|  |  |  |     """Parse the text written in green.
 | 
					
						
							|  |  |  | 
 | 
					
						
							|  |  |  |     Possible formats: | 
					
						
							|  |  |  |     * "{authors} - {journal}, {year} - {publisher}" | 
					
						
							|  |  |  |     * "{authors} - {year} - {publisher}" | 
					
						
							|  |  |  |     * "{authors} - {publisher}" | 
					
						
							|  |  |  |     """
 | 
					
						
							|  |  |  |     if text is None or text == "": | 
					
						
							|  |  |  |         return None, None, None, None | 
					
						
							|  |  |  | 
 | 
					
						
							|  |  |  |     s_text = text.split(' - ') | 
					
						
							|  |  |  |     authors = s_text[0].split(', ') | 
					
						
							|  |  |  |     publisher = s_text[-1] | 
					
						
							|  |  |  |     if len(s_text) != 3: | 
					
						
							|  |  |  |         return authors, None, publisher, None | 
					
						
							|  |  |  | 
 | 
					
						
							|  |  |  |     # the format is "{authors} - {journal}, {year} - {publisher}" or "{authors} - {year} - {publisher}" | 
					
						
							|  |  |  |     # get journal and year | 
					
						
							|  |  |  |     journal_year = s_text[1].split(', ') | 
					
						
							|  |  |  |     # journal is optional and may contains some coma | 
					
						
							|  |  |  |     if len(journal_year) > 1: | 
					
						
							|  |  |  |         journal = ', '.join(journal_year[0:-1]) | 
					
						
							|  |  |  |         if journal == '…': | 
					
						
							|  |  |  |             journal = None | 
					
						
							|  |  |  |     else: | 
					
						
							|  |  |  |         journal = None | 
					
						
							|  |  |  |     # year | 
					
						
							|  |  |  |     year = journal_year[-1] | 
					
						
							|  |  |  |     try: | 
					
						
							|  |  |  |         publishedDate = datetime.strptime(year.strip(), '%Y') | 
					
						
							|  |  |  |     except ValueError: | 
					
						
							|  |  |  |         publishedDate = None | 
					
						
							|  |  |  |     return authors, journal, publisher, publishedDate | 
					
						
							|  |  |  | 
 | 
					
						
							|  |  |  | 
 | 
					
						
							|  |  |  | def response(resp):  # pylint: disable=too-many-locals | 
					
						
							| 
									
										
										
										
											2022-12-04 22:57:22 +01:00
										 |  |  |     """Parse response from Google Scholar""" | 
					
						
							| 
									
										
										
										
											2021-03-01 15:02:11 +01:00
										 |  |  |     results = [] | 
					
						
							|  |  |  | 
 | 
					
						
							|  |  |  |     # convert the text to dom | 
					
						
							|  |  |  |     dom = html.fromstring(resp.text) | 
					
						
							| 
									
										
										
										
											2022-12-04 22:57:22 +01:00
										 |  |  |     detect_google_captcha(dom) | 
					
						
							| 
									
										
										
										
											2021-03-01 15:02:11 +01:00
										 |  |  | 
 | 
					
						
							|  |  |  |     # parse results | 
					
						
							| 
									
										
										
										
											2022-12-04 22:57:22 +01:00
										 |  |  |     for result in eval_xpath_list(dom, '//div[@data-rp]'): | 
					
						
							| 
									
										
										
										
											2021-03-01 15:02:11 +01:00
										 |  |  | 
 | 
					
						
							| 
									
										
										
										
											2022-08-26 18:10:12 +02:00
										 |  |  |         title = extract_text(eval_xpath(result, './/h3[1]//a')) | 
					
						
							| 
									
										
										
										
											2021-03-01 15:02:11 +01:00
										 |  |  | 
 | 
					
						
							|  |  |  |         if not title: | 
					
						
							|  |  |  |             # this is a [ZITATION] block | 
					
						
							|  |  |  |             continue | 
					
						
							|  |  |  | 
 | 
					
						
							| 
									
										
										
										
											2022-12-04 22:57:22 +01:00
										 |  |  |         pub_type = extract_text(eval_xpath(result, './/span[@class="gs_ctg2"]')) | 
					
						
							| 
									
										
										
										
											2021-03-01 15:02:11 +01:00
										 |  |  |         if pub_type: | 
					
						
							| 
									
										
										
										
											2022-08-26 18:10:12 +02:00
										 |  |  |             pub_type = pub_type[1:-1].lower() | 
					
						
							|  |  |  | 
 | 
					
						
							|  |  |  |         url = eval_xpath_getindex(result, './/h3[1]//a/@href', 0) | 
					
						
							|  |  |  |         content = extract_text(eval_xpath(result, './/div[@class="gs_rs"]')) | 
					
						
							|  |  |  |         authors, journal, publisher, publishedDate = parse_gs_a( | 
					
						
							|  |  |  |             extract_text(eval_xpath(result, './/div[@class="gs_a"]')) | 
					
						
							|  |  |  |         ) | 
					
						
							|  |  |  |         if publisher in url: | 
					
						
							|  |  |  |             publisher = None | 
					
						
							|  |  |  | 
 | 
					
						
							|  |  |  |         # cited by | 
					
						
							|  |  |  |         comments = extract_text(eval_xpath(result, './/div[@class="gs_fl"]/a[starts-with(@href,"/scholar?cites=")]')) | 
					
						
							|  |  |  | 
 | 
					
						
							|  |  |  |         # link to the html or pdf document | 
					
						
							|  |  |  |         html_url = None | 
					
						
							|  |  |  |         pdf_url = None | 
					
						
							|  |  |  |         doc_url = eval_xpath_getindex(result, './/div[@class="gs_or_ggsm"]/a/@href', 0, default=None) | 
					
						
							|  |  |  |         doc_type = extract_text(eval_xpath(result, './/span[@class="gs_ctg2"]')) | 
					
						
							|  |  |  |         if doc_type == "[PDF]": | 
					
						
							|  |  |  |             pdf_url = doc_url | 
					
						
							|  |  |  |         else: | 
					
						
							|  |  |  |             html_url = doc_url | 
					
						
							| 
									
										
										
										
											2021-03-01 15:02:11 +01:00
										 |  |  | 
 | 
					
						
							| 
									
										
										
										
											2021-12-27 09:26:22 +01:00
										 |  |  |         results.append( | 
					
						
							|  |  |  |             { | 
					
						
							| 
									
										
										
										
											2022-08-26 18:10:12 +02:00
										 |  |  |                 'template': 'paper.html', | 
					
						
							|  |  |  |                 'type': pub_type, | 
					
						
							| 
									
										
										
										
											2021-12-27 09:26:22 +01:00
										 |  |  |                 'url': url, | 
					
						
							|  |  |  |                 'title': title, | 
					
						
							| 
									
										
										
										
											2022-08-26 18:10:12 +02:00
										 |  |  |                 'authors': authors, | 
					
						
							|  |  |  |                 'publisher': publisher, | 
					
						
							|  |  |  |                 'journal': journal, | 
					
						
							|  |  |  |                 'publishedDate': publishedDate, | 
					
						
							| 
									
										
										
										
											2021-12-27 09:26:22 +01:00
										 |  |  |                 'content': content, | 
					
						
							| 
									
										
										
										
											2022-08-26 18:10:12 +02:00
										 |  |  |                 'comments': comments, | 
					
						
							|  |  |  |                 'html_url': html_url, | 
					
						
							|  |  |  |                 'pdf_url': pdf_url, | 
					
						
							| 
									
										
										
										
											2021-12-27 09:26:22 +01:00
										 |  |  |             } | 
					
						
							|  |  |  |         ) | 
					
						
							| 
									
										
										
										
											2021-03-01 15:02:11 +01:00
										 |  |  | 
 | 
					
						
							|  |  |  |     # parse suggestion | 
					
						
							|  |  |  |     for suggestion in eval_xpath(dom, '//div[contains(@class, "gs_qsuggest_wrap")]//li//a'): | 
					
						
							|  |  |  |         # append suggestion | 
					
						
							|  |  |  |         results.append({'suggestion': extract_text(suggestion)}) | 
					
						
							|  |  |  | 
 | 
					
						
							|  |  |  |     for correction in eval_xpath(dom, '//div[@class="gs_r gs_pda"]/a'): | 
					
						
							|  |  |  |         results.append({'correction': extract_text(correction)}) | 
					
						
							|  |  |  | 
 | 
					
						
							|  |  |  |     return results |