SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2005.10107 · ACL · 2020

Examining the State-of-the-Art in News Timeline Summarization

Demian Ghalandari, Georgiana Ifrim

arXiv · PDF · Open in the Atlas

Code that ran

5 of 8 samples (functions and classes) lifted from this paper's linked repositories executed on a synthesized input in Syntology's jail (3 of them constructed an object rather than computing a result; 1 honoured the contract we drafted, 0 violated it and 4 have no contract verdict; 0 failures of our instrument; 3 no recorded run), measured 2026-11-22; repository links from the Papers with Code archive attributed under CC BY-SA, no affiliation; bodies inline only under a permissive licence as GitHub detects it. Not a reproduction of the paper's results.

"Ran" means the function executed on a synthesized input and returned a value. It is not a reproduction of the paper's results. This record, dated, as JSON: /api/ran/2005.10107.json · badge: /api/ran/2005.10107.svg

Each repository's role is a label: a classifier's output recorded on the link, or, where no role is recorded, what the Papers with Code archive marked. The line under it names its basis. Neither says who wrote the code.

RepositoryRoleRan
complementizer/news-tls named in the paper
Papers with Code archive
5 ran · 3 no recorded run
FunctionStatusWhere it lives
CentroidOpt Ran
ran on a synthesized input · constructed an object; no result was computed
complementizer/news-tls/news_tls/datewise.py
code served (permissive licence) · get_code("e91de713e9c8a055")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned an instance of its own class, CentroidOpt: the run constructed the object and did not call it, so no result was computed. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import numpy as np
from scipy import sparse
from sklearn.feature_extraction.text import TfidfVectorizer

class Sentence:
    def __init__(self, raw):
        self.raw = raw

# Create test sentences
sents = [
    Sentence("The quick brown fox jumps over the lazy dog"),
    Sentence("A fast brown fox leaps over a lazy dog"),
    Sentence("The slow green turtle crawls under the busy cat"),
    Sentence("Machine learning is a subset of artificial intelligence"),
]

# Create a vectorizer
vectorizer = TfidfVectorizer(max_features=10)
vectorizer.fit([s.raw for s in sents])

# Create test data for CentroidOpt
k = 2
filter_func = lambda s: len(s.split()) > 3

ARGS = {
    'max_sim': 0.9999,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, scipy, sklearn — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type CentroidOpt — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a CentroidOpt output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
MentionCountDateRanker Ran
ran on a synthesized input
complementizer/news-tls/news_tls/datewise.py
code served (permissive licence) · get_code("245accd23f7b626a")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, class construct-then-call prompt, model not recorded) and returned list, ndim 1, dtype family float; no output shape is stated. The image it ran in was not recorded. If your copy returns the same type and ndim on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import numpy as np
from collections import defaultdict

# Stand-in classes to simulate the expected structure
class Sentence:
    def __init__(self, date=None):
        self.date = date

    def get_date(self):
        return self.date

class Article:
    def __init__(self, sentences):
        self.sentences = sentences

class Collection:
    def __init__(self, articles):
        self._articles = articles

    def articles(self):
        return self._articles

# Setup INIT_ARGS for MentionCountDateRanker
INIT_ARGS = {}

# Setup ARGS for rank_dates method
articles = [
    Article([Sentence(date="2023-01-01"), Sentence(date="2023-01-02")]),
    Article([Sentence(date="2023-01-01"), Sentence(date="2023-01-01"), Sentence(date="2023-01-03")]),
    Article([Sentence(date="2023-01-02"), Sentence(date="2023-01-04")])
]
collection = Collection(articles)
ARGS = {'collection': collection}
METHOD = 'rank_dates'
Setup block drafted by
a language model, class construct-then-call prompt, model not recorded
Imports
numpy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type list, ndim 1, dtype family float — a list output has no array shape of its own; none is stated
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: torch and numpy seeded to 0 by the harness immediately before the class was constructed from INIT_ARGS (fixture_check.py)
PM_Mean_SentenceCollector Ran
ran on a synthesized input · constructed an object; no result was computed
complementizer/news-tls/news_tls/datewise.py
code served (permissive licence) · get_code("82c3fe12d5eef499")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned an instance of its own class, PM_Mean_SentenceCollector: the run constructed the object and did not call it, so no result was computed. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import datetime
import collections
import numpy as np
from scipy import sparse
from sklearn.feature_extraction.text import TfidfVectorizer

# Stand-in classes for the data structures
class Sentence:
    def __init__(self, raw_text, mention_date=None):
        self.raw = raw_text
        self._mention_date = mention_date
    
    def get_date(self):
        return self._mention_date

class Article:
    def __init__(self, pub_time, title_sentence=None, sentences=None):
        self.time = pub_time
        self.title_sentence = title_sentence
        self.sentences = sentences if sentences is not None else []

class Collection:
    def __init__(self, articles_list):
        self._articles = articles_list
    
    def articles(self):
        return self._articles

class MockDateTime:
    def __init__(self, year, month, day):
        self.dt = datetime.datetime(year, month, day)
    
    def date(self):
        return self.dt.date()

# Create sample sentences
sent1 = Sentence("the quick brown fox", mention_date=datetime.date(2024, 1, 15))
sent2 = Sentence("jumps over the lazy dog", mention_date=datetime.date(2024, 1, 15))
sent3 = Sentence("the stock market rose today", mention_date=None)
sent4 = Sentence("investors are optimistic", mention_date=None)

# Create sample articles
article1 = Article(
    pub_time=MockDateTime(2024, 1, 16),
    title_sentence=Sentence("Breaking News Title"),
    sentences=[sent3, sent4, sent1, sent2]
)

collection = Collection([article1])

# Create a simple vectorizer
vectorizer = TfidfVectorizer(max_features=10)
# Fit it with sample texts
sample_texts = [
    "the quick brown fox",
    "jumps over the lazy dog",
    "the stock market rose today",
    "investors are optimistic",
    "Breaking News Title"
]
vectorizer.fit(sample_texts)

# Create ranked dates for the second pass
ranked_dates = [datetime.date(2024, 1, 15), datetime.date(2024, 1, 16)]

ARGS = {
    "clip_sents": 5,
    "pub_end": 2,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, scipy, sklearn — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type PM_Mean_SentenceCollector — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a PM_Mean_SentenceCollector output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
Timeline Ran
ran on a synthesized input · constructed an object; no result was computed
complementizer/news-tls/news_tls/datewise.py
code served (permissive licence) · get_code("b8f913905948b26c")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned an instance of its own class, Timeline: the run constructed the object and did not call it, so no result was computed. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

from datetime import datetime, timedelta

# Create a list of (datetime, summary) tuples
# The Timeline constructor expects items to be tuples of (timestamp, summary)
# where timestamp has a .date() method and summary is iterable (contains sentences)

items = [
    (datetime(2023, 1, 1, 10, 30), ["First event happened today.", "It was significant."]),
    (datetime(2023, 1, 2, 14, 15), ["Second event occurred.", "More details here.", "Additional context."]),
    (datetime(2023, 1, 3, 9, 0), ["Third event summary."]),
]

ARGS = {
    "items": items
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
none outside the standard library — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Timeline — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a Timeline output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
detect_knee_point Ran
ran, honoured the contract we drafted
complementizer/news-tls/news_tls/datewise.py
code served (permissive licence) · get_code("03b0dc69a5925947")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned int64, ndim 0, dtype family int, shape () as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import numpy as np

# The function detect_knee_point takes a 1D array-like of values
# and finds the "knee point" - the point with maximum perpendicular distance
# from the line connecting the first and last points.

# Create a simple set of values that shows a clear knee point
# For example, values that decrease rapidly then level off
values = np.array([100.0, 85.0, 75.0, 70.0, 68.0, 67.5, 67.2, 67.1, 67.0])

ARGS = {
    "values": values
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type int64, ndim 0, dtype family int, shape () — shape () as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
DateRanker Not yet run
unverified (harvested, no recorded run)
complementizer/news-tls/news_tls/datewise.py
code served (permissive licence) · get_code("cf8fe36d55b3528d")
How it ran

no invocation recorded.

DatewiseTimelineGenerator Not yet run
unverified (harvested, no recorded run)
complementizer/news-tls/news_tls/datewise.py
code served (permissive licence) · get_code("0fe52471ee37597b")
How it ran

no invocation recorded.

Summarizer Not yet run
unverified (harvested, no recorded run)
complementizer/news-tls/news_tls/datewise.py
code served (permissive licence) · get_code("38660adc18da47f5")
How it ran

no invocation recorded.

Also present in other papers' code

Of 8 samples lifted from this paper's linked repositories, 8 carry a function identity and 0 do not. Of the 8 distinct functions with an identity, 0 are also present in other papers whose code we hold (0 with at least one paper dated earlier, 0 the same month, 0 later) and 8 are only here among the 52,977 papers whose code Syntology holds (as of 2026-09-26).

Only here means: among the 52,977 papers whose code Syntology holds (51,887 with lifted samples, 1,090 through their assemblies only), as of 2026-09-26, no other paper's lifted samples or current assembly bricks contain an identical function body after normalisation.

Presence, not direction: an identical function body in two papers' repositories does not say who wrote it, which came from which, or that either paper's method rests on it. Matching is exact after normalisation, so a renamed or edited function is not matched. Dates are when papers appeared, not when code was written. Only .py files the harvester lifted are covered; notebooks are not read. A new-style arXiv id YYMM.NNNN dates a paper to the month 20YY-MM; otherwise Paper.year dates it to the year; otherwise it is undated. Dates are when the paper appeared, not when the code was written. Read 2026-11-22.

Repositories linked to this paper

Some links come from the archived Papers with Code dataset (CC BY-SA 4.0): attribution and licence.

Abstract

Previous work on automatic news timeline summarization (TLS) leaves an unclear picture about how this task can generally be approached and how well it is currently solved. This is mostly due to the focus on individual subtasks, such as date selection and date summarization, and to the previous lack of appropriate evaluation metrics for the full TLS task. In this paper, we compare different TLS strategies using appropriate evaluation frameworks, and propose a simple and effective combination of methods that improves over the stateof-the-art on all tested benchmarks. For a more robust evaluation, we also present a new TLS dataset, which is larger and spans longer time periods than previous datasets. The dataset will be made available at https://github. com/complementizer/news-tls.

For agents

The same record, over MCP at https://syntology.ai/mcp:

get_harvested_code_for_paper("2005.10107")
get_code_for_paper("2005.10107")
have("2005.10107")

The run record, dated, one paper per request, free:

curl https://syntology.ai/api/ran/2005.10107.json

A badge for a README (the split and the date, never a ratio):

[![Syntology run record](https://syntology.ai/api/ran/2005.10107.svg)](https://syntology.ai/paper/2005.10107)

Connect an agent — have() is free.