Abstract

It is the most important way for researchers to acquire academic progress via reading scientific papers, most of which are in PDF format. However, existing PDF Readers like Adobe Acrobat Reader and Foxit PDF Reader are usually only for reading by rendering PDF files as a whole, and do not consider the multi-granularity

Hammer PDF: An Intelligent PDF Reader for Scientific Papers
Sheng-Fu Wang
∗
Shu-Hang Liu
∗
{wangsf,liush}@bit.edu.cn
Beijing Institute of Technology
Haidian Distr., Beijing, China
Tian-Yi Che
ccty@bit.edu.cn
Beijing Institute of Technology
Haidian Distr., Beijing, China
Yi-Fan Lu
luyifan@bit.edu.cn
Beijing Institute of Technology
Haidian Distr., Beijing, China
Song-Xiao Yang
yangsongxiao0616@gmail.com
Beijing Institute of Technology
Haidian Distr., Beijing, China
Heyan Huang
hhy63@bit.edu.cn
Beijing Institute of Technology
Haidian Distr., Beijing, China
Xian-Ling Mao
†
maoxl@bit.edu.cn
Beijing Institute of Technology
Haidian Distr., Beijing, China
ABSTRACT
It is the most important way for researchers to acquire academic
progress via reading scientific papers, most of which are in PDF
format. However, existing PDF Readers like Adobe Acrobat Reader
and Foxit PDF Reader are usually only for reading by rendering
PDF files as a whole, and do not consider the multi-granularity
content understanding of a paper itself. Specifically, taking a paper
as a basic and separate unit, existing PDF Readers cannot access ex-
tended information about the paper, such as corresponding videos,
blogs and codes. Meanwhile, they cannot understand the academic
content of a paper, such as terms, authors, and citations. To solve
these problems, we introduce Hammer PDF, an intelligent PDF
Reader for scientific papers. Apart from basic reading functions,
Hammer PDF has the following four innovative features: (1) infor-
mation extraction ability, which can locate and mark spans like
terms and other entities; (2) information extension ability, which
can present relevant academic content of a paper, such as cita-
tions, references, codes, videos, blogs, etc; (3) built-in Hammer
Scholar, an academic search engine based on academic informa-
tion collected from major academic databases; (4) built-in Q&A
bot, which can find helpful conference information; The proposed
Hammer PDF Reader can help researchers, especially those study-
ing computer science, to improve the efficiency and experience of
reading scientific papers. We have released Hammer PDF, available
at https://pdf.hammerscholar.net/face.
CCS CONCEPTS
•Information systems→Web applications
;Search interfaces;
Information extraction.
∗
Both authors contributed equally to this research.
†
Corresponding author
Permission to make digital or hard copies of all or part of this work for personal or
classroom use is granted without fee provided that copies are not made or distributed
for profit or commercial advantage and that copies bear this notice and the full citation
on the first page. Copyrights for components of this work owned by others than ACM
must be honored. Abstracting with credit is permitted. To copy otherwise, or republish,
to post on servers or to redistribute to lists, requires prior specific permission and/or a
fee. Request permissions from permissions@acm.org.
CIKM ’22, October 17-22, 2022, Georgia, USA
©2022 Association for Computing Machinery.
ACM ISBN 978-1-4503-XXXX-X/18/06. . . $15.00
https://doi.org/XXXXXXX.XXXXXXX
KEYWORDS
PDF Reader, Literature Search, Information Extraction
ACM Reference Format:
Sheng-Fu Wang, Shu-Hang Liu, Tian-Yi Che, Yi-Fan Lu, Song-Xiao Yang,
Heyan Huang, and Xian-Ling Mao. 2022. Hammer PDF: An Intelligent
PDF Reader for Scientific Papers. InProceedings of 31st ACM International
Conference on Information and Knowledge Management (CIKM ’22).ACM,
New York, NY, USA, 5 pages. https://doi.org/XXXXXXX.XXXXXXX
1  INTRODUCTION
Nowadays, researchers have to spend considerable time on reading
scientific papers to keep abreast of all the latest developments
concerning their specialized fields. Most of the literature is in PDF
format, and there are many tools available on the market that
support reading PDF documents, such as Adobe Acrobat Reader
1
and Foxit PDF Reader
2
.
However, existing PDF Readers like Adobe Acrobat Reader and
Foxit PDF Reader are usually only for reading by rendering PDF
files as a whole, and do not consider the multi-granularity content
understanding of a paper itself. Specifically, taking a paper as a ba-
sic and separate unit, existing PDF Readers cannot access extended
information about the paper, such as corresponding videos, blogs
and codes. Meanwhile, they cannot understand the academic con-
tent of a paper, such as terms, authors and citations. For example,
when a paper is opened through Adobe Reader, researchers can
only read the paper itself, and cannot obtain the extended content
such as its corresponding videos, tutorial blogs and implementation
codes. If a researcher wants to know: "Where is the corresponding
code/video/blogs? What is the meaning of a word? What about
authors? Where is the full text of a reference?", he has to use other
tools such as browsers, translators, web search engines or scholar
search engines to find the answers, which is tedious and tend to
interrupt the reading process, and thus is very low efficient. Why
do not we use only one tool to accomplish all these functions?
Thus, to tackle these above problems, this paper will introduce
Hammer PDF, a multi-platform intelligent PDF Reader for scientific
papers, to improve the efficiency and experience of reading PDF
documents through machine learning and academic search. The
proposed Hammer PDF Reader is available for both web and desktop
1
https://www.adobe.com/acrobat/pdf-reader.html
2
https://www.foxit.com/pdf-reader/
arXiv:2204.02809v2  [cs.DL]  18 Jun 2022

CIKM ’22, October 17-22, 2022, Georgia, USAS.F. Wang and S.H. Liu, et al.
Conference
Q & A
Q&A bot
Basic
Reading
PDF
Render
Paper
Management
User
System
PDF
Parser
Information
Extension
Authors 
Blogs
Citations
Datasets
Codes
Slides
Videos
Source
...
External 
Knowledge Base
Quick
Jumping
Entity
Linking
Information
Extraction
Terms
Encyclopedia 
Articles
Paper
Web Crawler
DatabaseElasticsearch
Hammer Scholar
Tables
Figures
Equations
User
Figure 1: The architecture and workflow of Hammer PDF.
applications (Windows, macOS, and Linux). Hammer PDF has four
new features as follows:
•Information extraction ability, which can first get key spans
like terms, authors, and citations by information extraction
methods, and then mark these spans on the view panel, en-
abling users to interact directly with these spans.
•Information extension ability, which can present related aca-
demic information of a paper such as authors, citations, ref-
erences, codes, videos, and blogs.
•Built-in Hammer Scholar, which is an one-stop academic
search engine based on academic information collected from
major academic databases.
•Built-in Q&A bot, which can find useful conference informa-
tion, such as host place, host date, and impact factor.
2  RELATED WORK
Traditional PDF Readers like Adobe Acrobat Reader and Foxit PDF
Reader only support rendering, reading, and other basic functions.
Meanwhile, these PDF Readers do not perform any content analysis,
which fails to meet the intelligent needs of reading academic papers.
As a result, several new PDF Readers focusing on mining academic
value of scientific papers have come out recently.
ScholarPhi [3,4] argues that it is often difficult to read articles
when the information researchers need to understand them is scat-
tered across multiple paragraphs. Thus, ScholarPhi
3
identifies terms
and symbols in an article through fine-grained analysis and allows
users to access their definitions. ReadPaper
4
focuses on academic
communication through literature management and notes sharing,
along with simple paper search functions through coarse-grained
analysis. However, the academic database of ReadPaper is lacking
in content to meet complex reading demands.
3
https://www.semanticscholar.org/product/semantic-reader
4
https://readpaper.com/
Compared to existing PDF Readers, Hammer PDF, introduced in
this paper, is able to perform multi-granularity analysis of scientific
papers, providing academic enhancements as intelligent features.
3  OVERVIEW
In this section, we introduce the overall architecture and core fea-
tures of Hammer PDF. As shown in Figure 1, Hammer PDF features
basic reading as well as academic enhancements. Specifically, aca-
demic enhancements include information extraction, information
extension, academic search and conference Q&A. Section 4 provides
more details on academic enhancements.
Hammer PDF offers basic reading functions. Users can open
PDF documents via local file, URL address, or DOI (Digital Object
Identifier). Also, users can open documents from the search results
of the academic search engine Hammer Scholar
5
. When a document
is successfully opened, the interface will create a new tab called
the view panel, as shown in Figure 2 (a). The middle part of the
view panel displays the document itself, while the Basic Sidebar on
the left side provides text search and shows document information
including outline, thumbnail, and metadata. Also, on the right side
of the view panel, the Academic Sidebar presents additional content
for academic enhancements. Moreover, users can access translations
by directly selecting text within a document, which helps foreign
researchers quickly understand the content. As Figure 2 (f ) shows,
when some text is selected, a translation card is displayed near the
text and can be dragged as needed. In particular, users can change
the translation service in the settings.
Hammer PDF also has simple document management capabil-
ities, where users can open, bookmark, and delete one or more
document records. Furthermore, Hammer PDF is a multi-platform
PDF Reader available for web and desktop applications. In order to
satisfy the reading needs of as many users as possible, the interface
5
https://hammerscholar.net/

Hammer PDF: An Intelligent PDF Reader for Scientific PapersCIKM ’22, October 17-22, 2022, Georgia, USA
(c)
(b)
(d)
(e)(a)
(f)
Figure 2: The main function demonstration of Hammer PDF.
supports four languages, namely Simplified Chinese, Traditional
Chinese, English and Japanese.
4  INTELLIGENT FEATURES
In this section, we introduce four features of Hammer PDF for
academic enhancements, including information extraction, infor-
mation extension, academic search and conference Q&A.
4.1  Information Extraction
To structure documents, we use Grobid [5,7] to get the logical
structure of PDF documents. Grobid is a machine learning library
for extracting and parsing raw PDF files into structured documents,
with an F1-score of 0.89 in parsing references [8]. Then, we feed the
extracted title, abstract and body into the information extraction
model SpERT [2], a span-based joint entity and relation extraction
model. Specifically, we use SpERT to perform NER (Named Entity
Recognition) on the SciERC[6] dataset to obtain semantically rich
terms and their types, including Task, Method, Metric, Material,
Generic and Other. SpERT achieves an F1-score of 70.33% using
SciBERT[1] as a pre-trained language model for better results in
scientific papers.
With these terms in place, we identify the page number and
page location for each term. Next, we mark an interactive underline
mark where the term is located. Figure 2 (b) depicts that different
types of terms will be distinguished by different colored underlines.
The Academic Sidebar shows all terms in the document and their
respective contexts. When the user clicks on a term span, the sidebar
displays the translation and other locations for the same term.
We can also capture author spans and citation spans from the
structured document, and users are able to interact with these spans
in the view panel just like term spans, as shown in Figure 2 (b).
When an author span is clicked, the Academic Sidebar presents
the published works of the author. When a citation span is clicked,
the Academic Sidebar displays the details of the corresponding
reference.
4.2  Information Extension
Based on the academic database provided by Hammer Scholar,
which will be described in Section 4.3, we implement information
extension with multiple academic features. When the user opens
a document, we retrieve the academic database for a matching
paper according to the document’s structure information. Next, the
Academic Sidebar presents the information of the retrieved paper,
including the title, authors, abstract, citations, references, etc., as
depicted in Figure 2 (a). If the paper has related videos, tutorial
blogs, or implementation codes attached, they will be presented
as well. Besides, users can click on the name of an author or the
publication source to directly view its extended information.
We can perform information extension not only for papers but
also for terms. For example, when a term is selected, we fetch
the relevant encyclopedia from Wikipedia
6
and present it on the
Academic Sidebar. Specifically, for terms that cannot be matched
perfectly on Wikipedia, we offer several partial matches as an al-
ternative. In addition, figures, tables, and equations in the paper
support quick jumping, meaning that users can jump to the target
by clicking the corresponding button in the Academic Sidebar.
4.3  Built-in Academic Search
We collect academic resources from six literature databases in-
cluding arXiv
7
, ACL Anthology
8
and DBLP
9
. Academic resources
6
https://www.wikipedia.org/
7
https://arxiv.org/
8
https://aclanthology.org/
9
https://dblp.org/

CIKM ’22, October 17-22, 2022, Georgia, USAS.F. Wang and S.H. Liu, et al.
Figure 3: Search results for papers in Hammer Scholar
contain the title, author, publication date, publisher, DOI, abstract,
etc. Furthermore, we also collect presentations, blogs, videos, codes,
and other extended resources for information extension. Over these
resources, we build an academic search engine, named Hammer
Scholar.
Hammer Scholar has a separate interface that provides both
paper search and video search. Take paper search as an example,
the returned results after entering the keyword "dialog" are shown
in Figure 3. Apart from filtering and sorting current search results,
users can also pick a paper of interest and read it directly in the
view panel by click "Open" button in a result, eliminating the need
to find and upload the document.
4.4  Built-in Q&A Bot
Academic Q&A serves as a supplement to information extension
and can answer questions related to conferences. We design sev-
eral conference-related questions, such as host date, host place,
deadline, conference level, impact factor, etc. After the user asks a
question, academic Q&A identifies the question’s intent. If the ques-
tion belongs to a conference recommendation, then the Q&A model
returns the answer in natural language. If the question belongs to
an academic search, the interface will jump to Hammer Scholar and
search directly with the entered keywords. As depicted in Figure
4, when the user asks "What is the IF of TKDE?", the Q&A model
is able to reply the correct impact factor of the academic journal,
while when asking "what conferences have been held in May 2022?",
it returns a list of eligible conferences. When the mouse hovers
over the name of a conference, the user can view the conference’s
details.
To take previous conversations into account during the current
conversation, we also design a multi-round conversation logic. For
instance, if a user asks “What is the deadline of ACL” and then asks
“Where is it held”, the Q&A model is able to correctly say where
ACL is held. Notably, academic Q&A supports both Chinese and
English to serve more users.
5  DEMONSTRATION
We demonstrate several core features of Hammer PDF through
a comprehensive use case while using the figures from Section
4 due to space constraints. First, users have 3 ways to open an
academic article in PDF format: (1) upload a local file; (2) enter a
Figure 4: Conference information query in academic Q&A
URL address of a paper; (3) click "Open" button in returned results
from built-in Hammer Scholar search engine, as shown in Figure
3. Then, users can browse the document in the view panel and
check the document information in the Basic Sidebar, as seen in
Figure 2 (a). After that, the document is structured and parsed for
academic enhancements. Once the document has been processed,
the Academic Sidebar displays extended information about the
paper, and users can switch the top navigation to view all terms
within the article, as shown in Figure 2 (b). While reading, users can
click on an author mark to see the author’s detailed information
(Figure 2 (c)), translate the selected text (Figure 2 (f )), or click on a
citation mark to see the reference’s details, as depicted in Figure 2
(d). Furthermore, users are also able to click on the "Open" button
shown in Figure 2 (d) to access the full text of the reference of
interest directly in a new view panel.
Users who need to find a specific academic resource can use the
built-in academic search engine to search for it. As illustrated in
Figure 3, for a certain search result, users can click on the title to
view its details. Users can also chat with the conference Q&A bot to
obtain detail information about the academic journals or academic
conference, as depicted in Figure 4.
6  CONCLUSION
In this paper, we introduce Hammer PDF, a novel multi-platform
intelligent PDF Reader for scientific papers. Hammer PDF attempts
to meet the growing intelligent needs of researchers during read-
ing PDF documents. In addition to basic reading functions, we
improve reading efficiency through information extraction, infor-
mation extension, academic search, and conference Q&A. With lots
of academic data, Hammer PDF can boost researchers’ experience
when reading scientific papers.
The introduction video and usage examples can be found on
https://pdf.hammerscholar.net/face. We recommend using browsers
based on Chromium
10
to visit this site. Users can download desktop
applications (Windows, macOS, and Linux) and submit feedback
on https://github.com/HammerPDF/Smart-Scientific-Reader.
10
https://www.chromium.org/chromium-projects/

Hammer PDF: An Intelligent PDF Reader for Scientific PapersCIKM ’22, October 17-22, 2022, Georgia, USA
REFERENCES
[1]Iz Beltagy, Kyle Lo, and Arman Cohan. 2019.  SciBERT: A Pretrained Language
Model for Scientific Text. InProceedings of the 2019 Conference on Empirical Methods
in Natural Language Processing and the 9th International Joint Conference on Natural
Language Processing (EMNLP-IJCNLP). 3615–3620.
[2]Markus Eberts and Adrian Ulges. 2020.  Span-Based Joint Entity and Relation
Extraction with Transformer Pre-Training. InECAI 2020. IOS Press, 2006–2013.
[3]Andrew Head, Kyle Lo, Dongyeop Kang, Raymond Fok, Sam Skjonsberg, Daniel S
Weld, and Marti A Hearst. 2021. Augmenting scientific papers with just-in-time,
position-sensitive definitions of terms and symbols. InProceedings of the 2021 CHI
Conference on Human Factors in Computing Systems. 1–18.
[4]
Dongyeop Kang, Andrew Head, Risham Sidhu, Kyle Lo, Daniel S Weld, and Marti A
Hearst. 2020.   Document-Level Definition Detection in Scholarly Documents:
Existing Models, Error Analyses, and Future Directions. InProceedings of the First
Workshop on Scholarly Document Processing. 196–206.
[5]Patrice Lopez. 2009. GROBID: Combining automatic bibliographic data recognition
and term extraction for scholarship publications. InInternational conference on
theory and practice of digital libraries. Springer, 473–474.
[6]Yi Luan, Luheng He, Mari Ostendorf, and Hannaneh Hajishirzi. 2018. Multi-Task
Identification of Entities, Relations, and Coreference for Scientific Knowledge
Graph Construction. InProceedings of the 2018 Conference on Empirical Methods in
Natural Language Processing. 3219–3232.
[7]
Laurent Romary and Patrice Lopez. 2015.  Grobid-information extraction from
scientific publications.ERCIM News100 (2015).
[8]Dominika Tkaczyk, Andrew Collins, Paraic Sheridan, and Joeran Beel. 2018. Ma-
chine learning vs. rules and out-of-the-box vs. retrained: An evaluation of open-
source bibliographic reference and citation parsers. InProceedings of the 18th
ACM/IEEE on joint conference on digital libraries. 99–108.