Paper scoring system (#67)

breuleux · web-flow · commit f1fcaa324152 · 2025-07-10T14:43:42.000-04:00
* Add focus system for scoring
* Add mostly_latin and test_utils
* Increase default limit for OpenReview scraper
* Split/lowercase institutions before scoring
* Fix imports of PaperInfo, update dependencies
diff --git a/src/paperoni/__main__.py b/src/paperoni/__main__.py
@@ -8,9 +8,9 @@
 from gifnoc import add_overlay, cli
 from serieux import Auto, Registered, Tagged, TaggedUnion, serialize, singleton
 
-from .config import discoverers
-from .discovery.base import PaperInfo
+from .config import config, discoverers
 from .display import display, terminal_width
+from .model import PaperInfo
 
 
 class Formatter(Registered):
@@ -60,8 +60,14 @@ class Discover:
         # Output format
         format: Formatter = TerminalFormatter
 
+        # Top n entries
+        top: int = 0
+
         def run(self):
-            self.format(self.command())
+            papers = self.command()
+            if self.top:
+                papers = config.focuses.top(n=self.top, pinfos=papers)
+            self.format(papers)
 
     @dataclass
     class PaperoniInterface:
diff --git a/src/paperoni/config.py b/src/paperoni/config.py
@@ -9,6 +9,7 @@
 from serieux import TaggedSubclass
 
 from .discovery.base import Discoverer
+from .model.focus import Focuses
 
 
 @dataclass
@@ -32,6 +33,7 @@ class PaperoniConfig:
     cache_path: Path = None
     requests: RequesterConfig = field(default_factory=RequesterConfig)
     discovery: dict[str, TaggedSubclass[Discoverer]] = field(default_factory=dict)
+    focuses: Focuses = field(default_factory=Focuses)
 
 
 config = gifnoc.define(
diff --git a/src/paperoni/discovery/base.py b/src/paperoni/discovery/base.py
@@ -1,7 +1,4 @@
 from dataclasses import dataclass
-from datetime import datetime
-
-from ..model.classes import Paper
 
 
 @dataclass
@@ -11,11 +8,3 @@ class Discoverer:
 
 class QueryError(Exception):
     pass
-
-
-@dataclass(kw_only=True)
-class PaperInfo:
-    paper: Paper
-    key: str
-    update_key: str = None
-    acquired: datetime
diff --git a/src/paperoni/discovery/jmlr.py b/src/paperoni/discovery/jmlr.py
@@ -11,12 +11,13 @@
     Link,
     Paper,
     PaperAuthor,
+    PaperInfo,
     Release,
     Venue,
     VenueType,
 )
 from ..utils import asciiify
-from .base import Discoverer, PaperInfo
+from .base import Discoverer
 
 
 class JMLR(Discoverer):
diff --git a/src/paperoni/discovery/miniconf.py b/src/paperoni/discovery/miniconf.py
@@ -12,12 +12,13 @@
     Link,
     Paper,
     PaperAuthor,
+    PaperInfo,
     Release,
     Topic,
     Venue,
     VenueType,
 )
-from .base import Discoverer, PaperInfo
+from .base import Discoverer
 
 conference_urls = {
     "neurips": "neurips.cc",
diff --git a/src/paperoni/discovery/openalex.py b/src/paperoni/discovery/openalex.py
@@ -12,13 +12,14 @@
     Link,
     Paper,
     PaperAuthor,
+    PaperInfo,
     Release,
     Topic,
     Venue,
     VenueType,
 )
 from ..utils import QueryError, link_generators as LINK_GENERATORS
-from .base import Discoverer, PaperInfo
+from .base import Discoverer
 
 # https://docs.openalex.org/api-entities/institutions/institution-object#type
 INSTITUTION_CATEGORY_MAPPING = {
diff --git a/src/paperoni/discovery/openreview.py b/src/paperoni/discovery/openreview.py
@@ -13,12 +13,13 @@
     Link,
     Paper,
     PaperAuthor,
+    PaperInfo,
     Release,
     Topic,
     Venue,
     VenueType,
 )
-from .base import Discoverer, PaperInfo
+from .base import Discoverer
 
 
 def extract_date(txt: str) -> dict | None:
@@ -511,7 +512,7 @@ def query(
         # Block size for fetching results
         block_size: int = 100,
         # Maximum number of results to return
-        limit: int = 10000,
+        limit: int = 100000,
     ):
         """Query OpenReview"""
         for api_version in self.api_versions:
diff --git a/src/paperoni/discovery/pmlr.py b/src/paperoni/discovery/pmlr.py
@@ -10,12 +10,13 @@
     Link,
     Paper,
     PaperAuthor,
+    PaperInfo,
     Release,
     Venue,
     VenueType,
 )
 from ..utils import asciiify
-from .base import Discoverer, PaperInfo
+from .base import Discoverer
 
 
 def parse_paper(entry):
diff --git a/src/paperoni/discovery/semantic_scholar.py b/src/paperoni/discovery/semantic_scholar.py
@@ -9,12 +9,13 @@
     Link,
     Paper,
     PaperAuthor,
+    PaperInfo,
     Release,
     Topic,
     Venue,
     VenueType,
 )
-from .base import Discoverer, PaperInfo, QueryError
+from .base import Discoverer, QueryError
 
 external_ids_mapping = {
     "pubmedcentral": "pmc",
diff --git a/src/paperoni/model/__init__.py b/src/paperoni/model/__init__.py
@@ -7,6 +7,7 @@
     Link,
     Paper,
     PaperAuthor,
+    PaperInfo,
     Release,
     Topic,
     Venue,
@@ -22,6 +23,7 @@
     "Link",
     "Paper",
     "PaperAuthor",
+    "PaperInfo",
     "Release",
     "Topic",
     "Venue",
diff --git a/src/paperoni/model/classes.py b/src/paperoni/model/classes.py
@@ -164,8 +164,8 @@ class Author:
 @dataclass
 class Institution:
     name: str
-    category: InstitutionCategory
-    aliases: list[str]
+    category: InstitutionCategory = InstitutionCategory.unknown
+    aliases: list[str] = field(default_factory=list)
 
 
 @dataclass
@@ -177,8 +177,8 @@ class Venue:
     date_precision: DatePrecision
     volume: str = None
     publisher: str = None
-    aliases: list[str]
-    links: list[Link]
+    aliases: list[str] = field(default_factory=list)
+    links: list[Link] = field(default_factory=list)
     open: bool = False
     peer_reviewed: bool = False
 
@@ -194,15 +194,23 @@ class Release:
 class PaperAuthor:
     author: Author
     display_name: str
-    affiliations: list[Institution]
+    affiliations: list[Institution] = field(default_factory=list)
 
 
 @dataclass
 class Paper:
     title: str
     abstract: str = None
-    authors: list[PaperAuthor]
-    releases: list[Release]
-    topics: list[Topic]
-    links: list[Link]
+    authors: list[PaperAuthor] = field(default_factory=list)
+    releases: list[Release] = field(default_factory=list)
+    topics: list[Topic] = field(default_factory=list)
+    links: list[Link] = field(default_factory=list)
     flags: list[Flag] = field(default_factory=list)
+
+
+@dataclass(kw_only=True)
+class PaperInfo:
+    paper: Paper
+    key: str
+    update_key: str = None
+    acquired: datetime = field(default_factory=datetime.now)
diff --git a/src/paperoni/model/focus.py b/src/paperoni/model/focus.py
@@ -0,0 +1,104 @@
+import re
+from bisect import bisect_left
+from dataclasses import dataclass, field, replace
+
+from ovld import ovld
+
+from ..utils import mostly_latin
+from .classes import Paper, PaperAuthor, PaperInfo
+
+
+@dataclass
+class Focus:
+    type: str
+    name: str
+    score: float
+    drive_discovery: bool = False
+
+    @classmethod
+    def encode(cls, f):
+        drive_prefix = "!" if f.drive_discovery else ""
+        return f"{drive_prefix}{f.type}::{f.name}::{f.score}"
+
+    @classmethod
+    def decode(cls, s: str):
+        drive_discovery = s.startswith("!")
+        if drive_discovery:
+            s = s[1:]
+        type_name, name, score_str = [x.strip() for x in s.split("::")]
+        score = float(score_str)
+        return cls(
+            type=type_name, name=name, score=score, drive_discovery=drive_discovery
+        )
+
+    @classmethod
+    def serieux_model(cls, call_next):
+        return replace(
+            call_next(cls),
+            from_string=cls.decode,
+            to_string=cls.encode,
+        )
+
+
+def combine(scores):
+    return sum(scores)
+
+
+@dataclass
+class Focuses:
+    focuses: list[Focus] = field(default_factory=list)
+
+    def __post_init__(self):
+        self.score_index = {(f.type, f.name.lower()): f.score for f in self.focuses}
+
+    @classmethod
+    def serieux_deserialize(cls, obj, ctx, call_next):
+        return cls(call_next(list[Focus], obj, ctx))
+
+    @classmethod
+    def serieux_serialize(cls, obj, ctx, call_next):
+        return call_next(list[Focus], obj.focuses, ctx)
+
+    @ovld
+    def score(self, p: PaperInfo):
+        return self.score(p.paper)
+
+    @ovld
+    def score(self, p: Paper):
+        if not mostly_latin(p.title):
+            return 0.0
+        scores = [self.score(author) for author in p.authors]
+        return combine(scores)
+
+    @ovld
+    def score(self, p: PaperAuthor):
+        name_score = self.score_index.get(("author", p.display_name.lower()), 0.0)
+        iscores = [
+            self.score_index.get(("institution", name.lower()), 0.0)
+            for aff in p.affiliations
+            for name in re.split(r" *[,;/-] *", aff.name)
+        ]
+        return combine([name_score, *iscores])
+
+    def top(self, pinfos, n):
+        t = Top(n, self.score)
+        t.add_all(pinfos)
+        return t
+
+
+class Top(list):
+    def __init__(self, n, key):
+        self.n = n
+        self.key = lambda x: -key(x)
+
+    def add(self, x):
+        ins = bisect_left(self, self.key(x), key=self.key)
+        self.insert(ins, x)
+        del self[self.n :]
+
+    def add_all(self, elems):
+        for elem in elems:
+            self.add(elem)
+
+    def resort(self):
+        self.sort(key=self.key)
diff --git a/src/paperoni/utils.py b/src/paperoni/utils.py
@@ -86,3 +86,28 @@ def asciiify(s: str) -> str:
     norm = unicodedata.normalize("NFD", s)
     stripped = norm.encode("ASCII", "ignore")
     return stripped.decode("utf8")
+
+
+def mostly_latin(s: str, threshold: float = 0.9) -> bool:
+    """
+    Returns True if at least `threshold` (default 0.9) of the characters in the string
+    are ASCII or accented Latin characters.
+    """
+    ### LLM code
+    if not s:
+        return True
+    total = 0
+    good = 0
+    for c in s:
+        total += 1
+        o = ord(c)
+        if o < 128:
+            good += 1
+        else:
+            # Check if it's an accented Latin character
+            # Normalize and check if it decomposes to a Latin base
+            decomp = unicodedata.normalize("NFD", c)
+            base = decomp[0]
+            if "LATIN" in unicodedata.name(base, ""):
+                good += 1
+    return good / total >= threshold
diff --git a/tests/discovery/test_jmlr.py b/tests/discovery/test_jmlr.py
@@ -1,7 +1,7 @@
 from pytest_regressions.data_regression import DataRegressionFixture
 
-from paperoni.discovery.base import PaperInfo
 from paperoni.discovery.jmlr import JMLR
+from paperoni.model import PaperInfo
 
 from ..utils import check_papers
 
diff --git a/tests/discovery/test_miniconf.py b/tests/discovery/test_miniconf.py
@@ -3,8 +3,8 @@
 import pytest
 from pytest_regressions.data_regression import DataRegressionFixture
 
-from paperoni.discovery.base import PaperInfo
 from paperoni.discovery.miniconf import MiniConf, conference_urls
+from paperoni.model import PaperInfo
 
 from ..utils import check_papers, iter_affiliations
 
diff --git a/tests/discovery/test_openalex.py b/tests/discovery/test_openalex.py
@@ -1,8 +1,8 @@
 import pytest
 from pytest_regressions.data_regression import DataRegressionFixture
 
-from paperoni.discovery.base import PaperInfo
 from paperoni.discovery.openalex import OpenAlex, QueryError
+from paperoni.model import PaperInfo
 
 from ..utils import check_papers, filter_test_papers, iter_affiliations, split_on
 
diff --git a/tests/discovery/test_openreview.py b/tests/discovery/test_openreview.py
diff --git a/tests/discovery/test_pmlr.py b/tests/discovery/test_pmlr.py
diff --git a/tests/discovery/test_semantic_scholar.py b/tests/discovery/test_semantic_scholar.py
diff --git a/tests/model/__init__.py b/tests/model/__init__.py
diff --git a/tests/model/test_focus.py b/tests/model/test_focus.py
diff --git a/tests/test_utils.py b/tests/test_utils.py
diff --git a/tests/utils.py b/tests/utils.py
diff --git a/uv.lock b/uv.lock
