summaryrefslogtreecommitdiffstats
path: root/fatcat_scholar/search.py
blob: 611e185f06aeda4b5ae269f90509ee8e910d8427 (plain)
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
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
"""
Helpers to make elasticsearch queries.
"""

import sys
import datetime
from gettext import gettext
from typing import List, Optional, Any

import elasticsearch
from dynaconf import settings
from elasticsearch_dsl import Search, Q

# pytype: disable=import-error
from pydantic import BaseModel

# pytype: enable=import-error

# i18n note: the use of gettext below doesn't actually do the translation here,
# it just ensures that the strings are caught by babel for translation later


class FulltextQuery(BaseModel):
    q: Optional[str] = None
    limit: Optional[int] = None
    offset: Optional[int] = None
    filter_time: Optional[str] = None
    filter_type: Optional[str] = None
    filter_availability: Optional[str] = None
    sort_order: Optional[str] = None
    collapse_key: Optional[str] = None
    time_options: Any = {
        "label": gettext("Release Date"),
        "slug": "filter_time",
        "default": "all_time",
        "list": [
            {"label": gettext("All Time"), "slug": "all_time"},
            {"label": gettext("Past Week"), "slug": "past_week"},
            {"label": gettext("Past Year"), "slug": "past_year"},
            {"label": gettext("Since 2000"), "slug": "since_2000"},
            {"label": gettext("Before 1925"), "slug": "before_1925"},
        ],
    }
    type_options: Any = {
        "label": gettext("Resource Type"),
        "slug": "filter_type",
        "default": "papers",
        "list": [
            {"label": gettext("Papers"), "slug": "papers"},
            {"label": gettext("Reports"), "slug": "reports"},
            {"label": gettext("Datasets"), "slug": "datasets"},
            {"label": gettext("Everything"), "slug": "everything"},
        ],
    }
    availability_options: Any = {
        "label": gettext("Availability"),
        "slug": "filter_availability",
        "default": "fulltext",
        "list": [
            {"label": gettext("Fulltext"), "slug": "fulltext"},
            {"label": gettext("Metadata"), "slug": "everything"},
            {"label": gettext("Open Access"), "slug": "oa"},
        ],
    }
    sort_options: Any = {
        "label": gettext("Sort Order"),
        "slug": "sort_order",
        "default": "relevancy",
        "list": [
            {"label": gettext("Relevancy"), "slug": "relevancy"},
            {"label": gettext("Recent First"), "slug": "time_desc"},
            {"label": gettext("Oldest First"), "slug": "time_asc"},
        ],
    }


class FulltextHits(BaseModel):
    count_returned: int
    count_found: int
    offset: int
    limit: int
    deep_page_limit: int
    query_time_ms: int
    results: List[Any]


def do_fulltext_search(
    query: FulltextQuery, deep_page_limit: int = 2000
) -> FulltextHits:

    es_client = elasticsearch.Elasticsearch(settings.ELASTICSEARCH_BACKEND)
    search = Search(using=es_client, index=settings.ELASTICSEARCH_FULLTEXT_INDEX)

    # Convert raw DOIs to DOI queries
    if (
        query.q
        and len(query.q.split()) == 1
        and query.q.startswith("10.")
        and query.q.count("/") >= 1
    ):
        search = search.filter("terms", doi=query.q)
        query.q = "*"

    # type filters
    if query.filter_type == "papers" or query.filter_type is None:
        search = search.filter(
            "terms", type=["article-journal", "paper-conference", "chapter",]
        )
    elif query.filter_type == "reports":
        search = search.filter("terms", type=["report", "standard",])
    elif query.filter_type == "datasets":
        search = search.filter("terms", type=["dataset", "software",])
    elif query.filter_type == "everything":
        pass
    else:
        raise ValueError(
            f"Unknown 'filter_type' parameter value: '{query.filter_type}'"
        )

    # time filters
    if query.filter_time == "past_week":
        date_today = datetime.date.today()
        week_ago_date = str(date_today - datetime.timedelta(days=7))
        tomorrow_date = str(date_today + datetime.timedelta(days=1))
        search = search.filter("range", date=dict(gte=week_ago_date, lte=tomorrow_date))
    elif query.filter_time == "past_year":
        # (date in the past year) or (year is this year)
        # the later to catch papers which don't have release_date defined
        year_ago_date = str(datetime.date.today() - datetime.timedelta(days=365))
        this_year = datetime.date.today().year
        search = search.filter(
            Q("range", date=dict(gte=year_ago_date)) | Q("term", year=this_year)
        )
    elif query.filter_time == "since_2000":
        search = search.filter("range", year=dict(gte=2000))
    elif query.filter_time == "before_1925":
        search = search.filter("range", year=dict(lt=1925))
    elif query.filter_time == "all_time" or query.filter_time is None:
        pass
    else:
        raise ValueError(
            f"Unknown 'filter_time' parameter value: '{query.filter_time}'"
        )

    # availability filters
    if query.filter_availability == "oa":
        search = search.filter("term", tags="oa")
    elif query.filter_availability == "everything":
        pass
    elif query.filter_availability == "fulltext" or query.filter_availability is None:
        search = search.filter("terms", access_type=["wayback", "ia_file", "ia_sim"])
    else:
        raise ValueError(
            f"Unknown 'filter_availability' parameter value: '{query.filter_availability}'"
        )

    if query.collapse_key:
        search = search.filter("term", collapse_key=query.collapse_key)
    else:
        search = search.extra(
            collapse={
                "field": "collapse_key",
                "inner_hits": {"name": "more_pages", "size": 0,},
            }
        )

    # we combined several queries to improve scoring.

    # this query use the fancy built-in query string parser
    basic_fulltext = Q(
        "query_string",
        query=query.q,
        default_operator="AND",
        analyze_wildcard=True,
        allow_leading_wildcard=False,
        lenient=True,
        quote_field_suffix=".exact",
        fields=[
            "title^5",
            "biblio_all^3",
            "abstracts.body^2",
            "fulltext.body",
            "everything",
        ],
    )
    has_fulltext = Q("terms", access_type=["ia_sim", "ia_file", "wayback"],)
    poor_metadata = Q(
        "bool",
        should=[
            # if these fields aren't set, metadata is poor. The more that do
            # not exist, the stronger the signal.
            Q("bool", must_not=Q("exists", field="title")),
            Q("bool", must_not=Q("exists", field="year")),
            Q("bool", must_not=Q("exists", field="type")),
            Q("bool", must_not=Q("exists", field="stage")),
            Q("bool", must_not=Q("exists", field="biblio.container_ident")),
        ],
    )

    search = search.query(
        "boosting",
        positive=Q("bool", must=basic_fulltext, should=[has_fulltext],),
        negative=poor_metadata,
        negative_boost=0.5,
    )
    search = search.highlight(
        "abstracts.body",
        "fulltext.body",
        "fulltext.acknowledgment",
        "fulltext.annex",
        number_of_fragments=2,
        fragment_size=300,
        # TODO: this will fix highlight encoding, but requires ES 7.x
        # encoder="html",
    )

    # sort order
    if query.sort_order == "time_asc":
        search = search.sort("year", "date")
    elif query.sort_order == "time_desc":
        search = search.sort("-year", "-date")
    elif query.sort_order == "relevancy" or query.sort_order is None:
        pass
    else:
        raise ValueError(f"Unknown 'sort_order' parameter value: '{query.sort_order}'")

    # Sanity checks
    limit = min((int(query.limit or 15), 100))
    offset = max((int(query.offset or 0), 0))
    if offset > deep_page_limit:
        # Avoid deep paging problem.
        offset = deep_page_limit

    search = search[offset : (offset + limit)]

    try:
        resp = search.execute()
    except elasticsearch.exceptions.RequestError as e:
        # this is a "user" error
        print("elasticsearch 400: " + str(e.info), file=sys.stderr)
        if e.info.get("error", {}).get("root_cause", {}):
            raise ValueError(str(e.info["error"]["root_cause"][0].get("reason")))
        else:
            raise ValueError(str(e.info))
    except elasticsearch.exceptions.TransportError as e:
        # all other errors
        print("elasticsearch non-200 status code: {}".format(e.info), file=sys.stderr)
        raise IOError(str(e.info))

    # convert from objects to python dicts
    results = []
    for h in resp:
        r = h._d_
        # print(h.meta._d_)
        r["_highlights"] = []
        if "highlight" in dir(h.meta):
            highlights = h.meta.highlight._d_
            for k in highlights:
                r["_highlights"] += highlights[k]
        r["_collapsed"] = []
        r["_collapsed_count"] = 0
        if "inner_hits" in dir(h.meta):
            r["_collapsed_count"] = h.meta.inner_hits.more_pages.hits.total - 1
            for k in h.meta.inner_hits.more_pages:
                if k["key"] != r["key"]:
                    r["_collapsed"].append(k)
        results.append(r)

    for h in results:
        # Handle surrogate strings that elasticsearch returns sometimes,
        # probably due to mangled data processing in some pipeline.
        # "Crimes against Unicode"; production workaround
        for key in h:
            if type(h[key]) is str:
                h[key] = h[key].encode("utf8", "ignore").decode("utf8")
        # ensure collapse_key is a single value, not an array
        if type(h["collapse_key"]) == list:
            h["collapse_key"] = h["collapse_key"][0]

    count_found: int = int(resp.hits.total)
    count_returned = len(results)

    # if we grouped to less than a page of hits, update returned count
    if (not query.collapse_key) and offset == 0 and (count_returned < limit):
        count_found = count_returned

    return FulltextHits(
        count_returned=count_returned,
        count_found=count_found,
        offset=offset,
        limit=limit,
        deep_page_limit=deep_page_limit,
        query_time_ms=int(resp.took),
        results=results,
    )