With the advent of Research 2.0, the way research is conducted has significantly changed. New tools and methodologies have emerged and an increasing amount of research is conducted in networked communities including the use of social networking tools. Apart from the well-known social networks, smaller and tailored social networks for researchers have emerged that are geared towards the specific needs of researchers. As more and more potentially relevant information is being made available, many researchers feel the need for awareness support in order to cope with the available amount of data. In this article we introduce the PUSHPIN application that aims at supporting researchers' awareness of publications, peers and research trends. The application is based on an eResearch infrastructure that analyzes large corpora of scientific publications and combines the extracted data with the social interactions in an active social network.
In the early days, the Internet was mostly a top-down information distribution
system in which only few people provided information. Users of the Internet
merely consumed the information without being enabled to interact with or
create own information easily. With the rise of Web 2.0, Internet usage has
been revolutionized. It has enabled mankind to more easily participate in the
spread of information and the participation in global discourse [
The usage of Web 2.0 tools, practices and methodologies in the context of
scholarly communication has been recently labeled as Science 2.0 or Research
2.0 [
Thus, we especially consider the joint developing of ideas and the exchange of short texts, like in tweets or status updates as potentially relevant research information. Moreover, the use of social networks is considered as very relevant part of the modern research methodology. Despite the fact that Facebook has evolved to be the de-facto standard in social networking sites (SNS), there are several SNS that are tailored to the use by researchers and that help them in connecting to like-minded researcher, publications and other content.
Applications like Mendeley1, ResearchGate2, Academia.edu3 or iamResearcher4 compete with the top dog Facebook by providing features that cannot be found in the general-purpose social network. Mendeley for example focuses on the sharing and annotation of scientific documents in private or public groups. Moreover, it supports researchers in generating bibliographies and recommending publications that the research might be interested in.
However, the new way of conducting research, communicating research ideas and findings and sharing data also results in a very scattered network of potentially relevant information. Researchers are in urgent need of awareness support tools and techniques that provide detailed recommendations and hints for possible collaborators. Many of the existing approaches seem to be based on first-level metadata and collaborative filtering approaches only and this is where PUSHPIN (Supporting Scholarly Awareness in Publications and Social Networks ) will enhance the state-of-the-art. Through the application of in-depth publication and citation analysis combined with the immense power of the social graph, PUSHPIN aims to provide better awareness support for researchers than the existing tools.
In the following sections, we present our new application called PUSHPIN and its approach for awareness support for researchers (Section 2). In Section 3, we present the implementation details for PUSHPIN and present the underlying eResearch infrastructure. We also discuss the three user interfaces for web, mobile and tabletops that PUSHPIN provides for its users. Finally in Section 4, we give an outlook on future research opportunities and present our evaluation and public release plans.
2
The PUSHPIN approach for awareness support for researchers PUSHPIN is an ongoing research project at the University of Paderborn (Germany) that aims to provide awareness support for researchers through the integration of social networking and big data analysis features. While many features of the whole approach have already been implemented and can be used, other features are not yet realized and are currently under development.
In this section we give an introduction to how PUSHPIN will help researchers to become and stay aware of their connections to other researchers and publications. In particular we describe how the social layer and the available social networking features contribute to the overall awareness of researchers (Section 2.1) and discuss the power of email notifications to keep the users engaged to visit the platform (Section 2.6). In Section 2.3, we describe how the automatic analysis of big data sets of publications is supporting object-centered sociality in PUSHPIN and how it gives insight to the relations of people and objects in PUSHPIN. Moreover, we present visualizations (Section 2.5) and recommendations (Section 2.4) that support researchers’ awareness and discuss how we use mobile devices and interactive displays to access data in our ecosystem (Section 2.7). 2.1
To raise awareness of an idea and to create a circle of supporters of the same,
it is essential for any research idea to reach a wide audience. Social networking
makes it possible to connect to potential collaborators thereby supporting the
start of an incipient Research Network. Where social networking tools are often
based on the people element, on the other hand, social awareness tools tell us
a story using various data associated with people and helps us build a network
based on such data. Often, we also find social networks that assemble around
specific objects, which become the hub for social interactions [
The social layer of PUSHPIN aims to support users in creating an active
social network that is created by the users themselves through social interactions
and conscious activities. The other parts of PUSHPIN rather contribute to a
passive social network that is automatically generated by the system and that is
built based on abstract information and activities such as collaboratively writing
publications, working at the same institution or citing similar works [
in process for users and to support them to reuse their existing social profiles as login, we enable login via Facebook, Twitter and Mendeley. Moreover, PUSHPIN gets access to the respective social graphs and can recommend friends from the other social networks that already use PUSHPIN. User profile updates A user’s profile plays an important role in getting to know the user. To name a few, it consists of information about the user’s affiliations, research interests and research disciplines, which highlights the user’s research areas. This information has impact on the engagement in social networks as they reflect the personality of a user. Any changes to the user’s profile are presented to the followers of that user in their activity stream.
Following a user Users can follow other users to get an account of all their activities. The number of followers (users following the current user) and followees (users followed by the current user) are shown on the dashboard of a user as well as any user’s profile. The number of followers of a user can be taken as quantification of the popularity and networking efforts of that user on PUSHPIN.
Status updates, likes and comments Sharing status updates is a common construct in social networking applications which allow users to share their current thoughts or their work progress. In PUSHPIN, the status updates could be used not only for sharing ideas or current readings but also for requesting help or simply sharing some news. Also, all followers of the user can like and comment on a status message, which may eventually result in a discussion of the content shared. Moreover, if a user has connected other social media accounts to his PUSHPIN account, she can automatically share the status update with all of her other accounts.
Private messaging To support non-public information exchange, all users on PUSHPIN can exchange private messages with each other. Messages are stored in conversations that multiple users can be part of. Any member of a conversation can add additional users to the conversation and each user can leave a conversation at any time.
User’s activities When PUSHPIN users successfully sign in, they are redirected to their personal dashboard. A significant part of the dashboard consists of an activity stream, which is a sorted summary of activities. These activities consist of stories such as status updates of users, likes and comments on statuses, changes in profile information, users following and tagging other users, users uploading, bookmarking, rating and tagging publications, etc. In short, it tells stories of the users’ interaction with other users and publications. Users can only see updates of other users, whom they follow. Apart from the dashboard, users can also see activities of a particular user on their user profile. This kind of feature is common with most of the social networking platforms including Twitter and Facebook and hence, most of the users are already familiar with it. Uploading publications Since scientific publications are the central hub for object-centered sociality in PUSHPIN, users can upload publications to the service5. This may be done by selecting publication from the local computer and uploading them, or by connecting their Mendeley account to PUSHPIN. In the latter case, all the PDFs in the user’s Mendeley collections are automatically imported in the PUSHPIN infrastructure. All the publications that have been uploaded to the system, are then automatically analyzed and information is extracted from them (see Section 2.3 for a detailed description of this process).
Interacting with publications All the users have access to the dedicated profiles of all the publications in PUSHPIN. On the profile, users can rate the publication and share it on other social networking sites. Moreover, users can recommend the publication to other PUSHPIN users or send the recommendation via email. Finally, users can bookmark the publication and put it in one of their collections on PUSHPIN.
Tagging objects Social tagging is one of the most prominent features of Web
2.0 [
All scientific publications that are uploaded to the PUSHPIN infrastructure 6 are automatically analyzed according to several aspects. This automatic analysis represents a series of processing steps that are executed after a publication is uploaded the PUSHPIN system.
The first and foremost step taken is to check if the publication is already in
the publication corpus and/or if a full analysis has to be started. If this is not
the case, the uploaded publication is inserted into HBase7. After that, Storm8
is triggered for further analysis of the publication. This analysis by Storm
involves activating the metadata extraction and reference extraction modules to
obtain the metadata and references from the publication. The metadata,
being referred to, can be the title, the author(s) and their email addresses, the
authors’ institutions, abstract, and keywords. For each of the references that
have been cited in the publication, the reference extraction module looks for
title, author(s), year of publication and publication outlet. The two modules
5 Due to potential copyright infringements, we will only process the uploaded data in
order to extract metadata from the publications. We will not, however, allow the
public download of the PDFs shared with the PUSHPIN system.
6 Currently we only process articles in PDF format. In particular, we do not process
books or theses.
7 http://hbase.apache.org
8 http://storm-project.net
use GROBID9 and ParsCit10 as key software tools. If additional metadata in
BibTEX or PLoS XML format is available, the modules make use of this
information as well. The extracted data is then compared and combined to get
the most exact metadata (similar to our approach in [
A problem of modern science is the rising amount of plagiarism. In the digital age
it has become much easier to access scientific publications and to copy content. In
order to detect conscious or unconscious plagiarism we introduced algorithms to
PUSHPIN, which are capable of doing near-copy detection (NCD). NCD means
that correctly cited paragraphs will also be detected. To distinguish between
full-text quotes and plagiarism, additional algorithms have to be used to detect
plagiarism indicators. This could be done in future projects. The NCD algorithm
used in PUSHPIN are inspired by the fuzzy string similarity detection algorithm
described in [
Each uploaded paper first goes through initial text preprocessing steps before it can be analyzed by our NCD algorithm. These initial steps are used to remove irrelevant and uninteresting parts of the text and to make the different text better comparable: Text extraction The papers are uploaded as PDF files. From these files, the text, along with the information about its position in the PDF file are extracted. This gives the exact location of a copied text in the documents it appears in.
Text cleaning The extracted text contains – for the NCD algorithm – uninteresting information, like headers and footers of the document. These lines are removed and hyphenated words are joined again.
Language detection Some algorithms need to know the language of the text as they work with trained models that are specific for one language. Part-of-speech tagging The "Part-of-Speech" (POS) tagging determines the grammatical meaning of a word in a sentence. This information is necessary for detecting synonym groups of words later on. Moreover, POS tagging is also useful in combination with lemmatization for calculating word clouds. Lemmatization and stemming For comparing words in our NCD algorithm, it is necessary to bring all words to the principal form, which is the same for all tenses and plural and singular forms. Lemmatization transforms words in the principal form using a dictionary algorithm. This algorithm is expensive in time and memory but the results are real words, which also can be displayed in word clouds. Stemming is an algorithmic transformation of the input word that will transform it to the stem. The stem, however, does not
need to be a real word and thus should not be used in word clouds or the like but its calculation is very fast.
Number and stop word removal In this step, we remove unimportant elements from the text in order to reduce the complexity of the NCD algorithm computation.
Synonym detection Often, copiers try to conceal the copies by replacing words
with synonyms of the word. This makes it harder to detect certain parts of
a text as copied. This makes it necessary to detect synonym groups that a
given word belongs to and to check all synonyms of the word for potential
copies. In this step we make use of the WordNet project [
After the text preprocessing is finished, the NCD algorithm can calculate the
similarities between all sentences of the publication and the preprocessed
background corpus. This procedure is inspired by [
This computation needs very large computational power and produces a lot of similarity data. We rely on the Apache Hadoop framework to scale the computation to a cluster of computers (see Section 3 for a detailed inspection of the PUSHPIN eResearch infrastructure). 2.4
In PUSHPIN we use an ample number of recommender algorithms due to the following reasons: 1. The system has to take into consideration the networks that result from the extracted co-authorship information as well as the co-citation and bibliographic coupling data of publications. 2. For item-based recommendations, the system also has to employ the use of textual similarities, clustering results, author-assigned and extracted keywords as well as user tags. 3. Also, the system is capable of tracking user activity on the PUSHPIN web application, store the user activity, and based on these, be able to recommend resources (e.g., users who bookmarked publication X also bookmarked publication Y; mutual followers; you might also assign these tags to the resource because others did so; people who visited this resource also visited that resource).
To sum up, the recommender system takes into account all the above information for recommendation. In addition, the recommendations will be textual and visual, and also can be explained to the user. 2.5
Visualizations prove very useful in presenting and understanding large and
complex sets of data and mining for hidden patterns within them. They serve as a
very useful decision support tool in research networks and help researchers to
become and stay aware of large data sets [
Usage and statistical visualizations This category of visualizations will be prevalent throughout PUSHPIN. For researchers, there will be a simple chart depicting the development of followers, co-authors, publications, etc. Similarly, there will be charts for a publication how the number of citations and bookmarks developed over time. Besides, visualizations based on general statistical data like typical co-authorship network sizes, most referenced articles, top research disciplines, etc. will have a place in PUSHPIN. Trend-based visualizations This category will include trends using numbers as well as trends in usage of text over time. Trending citations, authors, topics and keywords will be visualized in appropriate manner.
Similarity-based visualizations Details of textual similarity between papers
and bibliographic coupling similarity between papers will be explored here.
Moreover, appropriate visualization of paragraphs that have been found
during the near-copy detection will be developed and provided in PUSHPIN.
Map-based visualizations Geo-spatial visualizations show us the
geographical location of researchers and institutions and help us understand the widely
spread co-authorship networks and the associations of different institutions
(inspired by the works of [
Co-authorship visualizations For a researcher, there will be a circular visualization with the researcher at center and his co-authors around him in circles. This give us a chance to explore the co-authors of this researcher. When a user explores a discipline, a research interest, an institution or a tag, there can be sets of co-authorship networks related to the explore query which may not be connected. Hence, we do not use a radial layout here, instead build a graph comprising of different networks(not connected) to show various sets effectively.
Besides the above categories, we will also have tag-based visualizations like word clouds, spark lines, etc. and also circle-based visualizations 2.6
As Fred Wilson points out “ if you want to drive retention and repeat usage [of
your service], there isn’t a better way to do it than email ” [
In our previous research we found that mobile access to research information,
together with context-awareness and push notification of relevant information is
very relevant for researchers overall awareness of their research networks [
The PUSHPINmobile application resembles a significant part of the features of the web application. Making use of specific mobile interface patterns such as dashboards and multitouch gestures, researchers are enabled to access all the information from the social layer and to engage in social interactions with their peers. Researchers are also able to view their own and other researchers’ profiles, search nearby researchers depending on their physical location, and also explore the different research disciplines, institutions and publications in the system. Moreover, researcher will also be able to tag other researchers and communicate with each other through private messages.
Beyond that, users of the mobile application will be enabled to authenticate and exchange data with the multitouch table application (PUSHPINMT ). Therefore, researchers can connect to PUSHPINMT using either Bluetooth or NFC. Additionally, the mobile application can bring up QR codes that can be scanned by the multitouch application. The QR codes can contain information about the researcher’ own or other researchers’ profile, institutions or publications. On PUSHPINMT , users will be able to explore their relations to other researchers and publications based on several scientometric measures. Moreover, they can explore the publications in PUSHPIN based on tags and other classifications. Finally, they can scan QR codes of any PUSHPIN object and get a virtual representation of the object on the tabletop. 3
PUSHPIN’s eResearch infrastructure implementation In this section we describe the technological underpinning of PUSHPIN’s eResearch and big data analysis infrastructure and relevant technologies we employ in the realization of the PUSHPIN user interfaces.
The Hadoop Distributed File System is an open source implementation of a
fault tolerant, self-healing, distributed filesystem for large datasets inspired by
the Google filesystem (GFS)[
Hadoop provides batch processing function, which perfectly scales with the number of nodes in a cluster. This functionality excellently supports parallelism to a wide range of algorithms especially in data mining and information retrieval.
In PUSHPIN, Hadoop is used for several algorithms, which need large computational performance and that process big data. Amongst others, these algorithms compute the similarity of texts, clusters the papers, builds recommender models or run near-copy detection algorithms. Moreover, we use Apache Mahout14 for the calculation of text-based similarities, text clustering, classification and recommender algorithms based on Hadoop MapReduce. 3.2
As described in Section 2.3, we perform several text preprocessing steps before a paper can be analyzed by the near-copy detection algorithm. The text extraction and thumbnail generation is done using Apache PDFBox15. Since many algorithms need to have knowledge about the language of a text, we use a Java-based language detection library16 for that. The Part-of-speech tagging is realized using Apache OpenNLP17. Stemming and lemmatization of the extracted texts is implemented on top of the Mate Tools natural language analysis toolkit18. Finally, we make use of Apache Lucene19 in the process of removing numbers and stop words that we consider as being not relevant for text similarities or near-copy detection. 3.3
During the metadata and reference extraction processes we are trying to accurately detect a publication’s title, author(s), contact information, like emails and 14 http://mahout.apache.org 15 http://pdfbox.apache.org 16 http://code.google.com/p/language-detection 17 http://opennlp.apache.org 18 http://code.google.com/p/mate-tools 19 http://lucene.apache.org address data as well as author-provided keywords and the publication’s abstract. Moreover, we are interested in the list of references and all the relevant data from each of the references. This metadata is extracted for different purposes, e.g., the attribution of publications to PUSHPIN users, the creation of co-authorship graphs, the calculation of recommendations and for detecting reference and research trends.
Once a publication has been uploaded to PUSHPIN and inserted into HBase,
the metadata and reference extraction modules get triggered by Storm. The
process involves triggering ParsCit and GROBID in parallel threads. GROBID
(GeneRatiOn of BIbliographic Data) employs the concept of Conditional
Random Fields (CRFs) for pattern recognition and data extraction [
Each tool does an independent metadata and reference extraction and the two results, obtained at the end, are then combined with potentially available other metadata like BibTEX data or PLoS XMLs. This merging is necessary as sometimes the metadata extracted from both tools differs, and also at times either of the tool misses out on some important metadata. If available, the data available in BibTEX or PLoS XML format are the most accurate source of information since they have been manually created by people knowledgeable of the publication. 3.4
In PUSHPIN, we use the Open Authorization (OAuth) protocol20 to allow users
to login to PUSHPIN using their Facebook, Twitter or Mendeley accounts.
OAuth “ is a security protocol that enables users to grant third-party access
to their web resources without sharing their passwords ” [
In PUSHPIN, we use provide a REST (REpresentational State Transfer) API (Application Programming Interface) to communicate between the frontends (web-based application, mobile application and multitouch table) and the Java backend. The frontend sends/requests data to the backend using the REST API, 20 http://oauth.net e.g., information about a certain resource such as a publication. The backend in turn returns a representation of the resource in JSON notation. The reasons for using REST (over other available web services such as SOAP) are that it is light-weight, simple, very popular among web applications and that it provides better performance and scalability. 3.6
PUSHPIN currently provides three user interfaces for its users. The web-based application serves as the main interface to our service and will be used by the average user. Moreover, we provide a mobile application for Android smartphones that allows the anytime-anywhere access to PUSHPIN’s main features. Finally, we also provide a multitouch application for tabletop-displays that supports users in exploring the PUSHPIN data in new ways.
Web-based application The web-based PUSHPIN front-end is a self-contained application and serves as the primary application to most of the users (see Figure 1). This application is written in PHP5 and builds on the state-of-the-art in HTML5 and CSS3 development. It also involves extensive use of JavaScript that enhances the user experience. Also, various Javascript frameworks are used for different visualizations.
Mobile application The PUSHPINmobile application is developed using the
Android 4 SDK and supports all smartphones running Android OS 4.0 and
higher. PUSHPINmobile currently provides users an interface to the social layer
of PUSHPIN and lets them flip through their activity stream, like and comment
entries and post new status updates. The application can scan QR codes of any
PUSHPIN object and present the data related to that object. Moreover, the
users can locate themselves and see relevant researchers around them.
Multitouch application The main purpose of the PUSHPINMT application
is to provide different interactions with the data in PUSHPIN. In [
Conclusion and future research opportunities In this paper we have introduced the PUSHPIN approach for awareness support in research networks. In PUSHPIN we combine the best of two worlds: classic features of Facebook-like social networking sites and those of innovative eResearch infrastructures. The integration of these features results in enhanced awareness support for researchers on both a social and a content layer. The recommender systems in PUSHPIN will not only recommend publications based on collaborative filtering but also on the actual content and reference data within the publications. Thus, PUSHPIN goes beyond the state-of-the-art and might help overcoming unwanted fragmentation in research networks and connecting researchers that otherwise would have stayed unknown to each other. In the coming months we will continue to improve the implementation of the analytical backend and further enhance the three user interfaces. We will invite selected users to an alpha test of the PUSHPIN web-based application in August and evaluate the existing features with them. The feedback on early versions of the software will help shaping the further development. We plan to release the system to public beta in early October 2012.