When writing a scientific paper one has always to deal with an plethora of literature on the topic. Reference management software was invented to support researchers in that regard: The tools are used to collect literature, manage references, and export bibliographies. They provide an efficient way to keep an overview of a large amount of literature. Numerous tools provide the opportunity to manage knowledge about references inside comments, notes, or tags. Researching is often a collaborative task demanding that literature management software should support collaboration. This includes sharing references and comments with other users or people who use another literature management software. Sharing comments with others may be beneficial 1) to ease understanding the paper itself and 2) to ease finding relevant papers, because indexable text is provided.
There are multiple resources on the web offering searching for literature such as "Google Scholar"3 or "The Collection of Computer Science Bibliographies"4. Many of them offer the functionality to import a reference into the preferred reference management software. However, they often provide incomplete or faulty reference entries. One exception is MathSciNet, where more than 20 persons take care of the quality of the references [8]. This quality assurance, however, is not implemented by all publishers.
A correct and complete entry is required for a correct reference list, which is a prerequisite for high-quality publications. Many programs for managing references provide a mechanism to detect missing required fields and highlight these entries to show the user that they are incomplete. However, this is not sufficient, because wrong information is not detected and it is tedious to find the correct missing information. The users have to check each reference entry manually to ensure correctness.
Hence, the goal of C R is to provide a cloud-based web application for collaborative reference management with to main features:
1. C R provides quality assurance by voting on bibliographical references to ensure complete and faultless references. 2. To support the cooperation of several people C R enables to post comments to literature at different levels of visibility. Sect. 2 presents related work on the field. Subsequently, Sect. 3 outlines the demonstration of C R followed by a description of the implementation (Sect. 4). Finally, Sect. 5 provides a discussion and an outlook on future work.
SoJa (Social JabRef [7]) introduces a social network among users. To establish a source of high-quality entries, for each topic, a user maintaining these entries has to be chosen in the community. There is no voting mechanism in place and comments can only be shared by embedding them into the BIBT E X file. Haase et al. [10] assume that there are BIBT E X databases with high-quality entries and that the issue is to identify duplicates and to find entries. To tackle these issues, they present the system Bibster. SharRef [22] focuses on sharing bibliographic data among groups and offers both a Web-based Client and a Java-based Desktop Client. Quality of entries is assured by having bibliography entries and automatically-updated shadows of them. There is no internal voting or commenting system in place. SocioBiblog [17] relies on the authors publishing their bibliographic data correctly on their homepages. There is no way presented on how to ensure quality of the resulting BIBT E X entries. BibSonomy [3] and its variant PUMA [4] offer to collect publications. It is possible to edit bibliographic data [6]. While there is a history function5, all edits are immediatly visible so there is no suggestion process as we propose. Academic search engines are surveyed by Ortega [16]. Additionally, there is OverCite [19] aggregating search results in a CiteSeer-like way. These tools offer search capabilities only and not a defined way to correct bibliographic entries. Beraka et al. [5] present a system for exchanging bibliographic information of scientific review and survey articles. Users can approve or disapprove bibliographic entries, but it is unclear how contradicting votes are treated. For presentation of surveys, SurVis [1] can be used. It is a read-only system without built-in functionalities of ensuring high-quality bibliographic data. Tkaczyk et al. [20] surveyed on reference parsers. They convert free reference text to a structured format. Thus, this is a way to get bibliographic data into a literature management system, but it is not ensured that the parsed data is of high quality itself. There is a movement on correctly citing software [18]. However, there is no quality control process proposed. Finally, we investigated 15 popular literature management systems6 and none of them offers 1) a voting system on bibliography entries and 2) comments with dedicated visibility.
After startup, the current prototype C R shows a table of all references (Fig. 1). References with a green check mark have been reviewed and marked as high-quality. A new reference can be input using a form-based editor or by uploading a BIBT E X file. A suggestion for improvements can be done using the entry editor. A user can click on "See suggestions for modifications" and a respective dialog is shown (Fig. 2). On the left side, there are the voting buttons. The number indicates the number of votes. After five positive votes by different users have been reached, the suggestion is merged.
In case a PDF is attached, comments on the PDF can be made (Fig. 3). A user can set the visibility of his own comments to public or private. "Public" denotes that each logged in user can see the comment and "private" denotes that the user exclusively can see the comment. This helps newcomers to research to make private notes and the more experienced researches to share their comments.
The architecture of C R including the used technologies is presented in Fig. 4. We implemented C R using Java and Angular. Regarding the storage, we decided to put the comments into a SQL database, placing the PDFs directly into the filesystem, and versioning the bibliography entries using git. Each suggestion becomes a new branch in git. In case a suggestion is accepted, the branch is merged. This way, we did not have to reinvent the whole branching and merging concept, but could rely on git's possibilities. The implementation is published at https://github.com/JabRef/cloudref/ and is offered as Docker image at https://hub.docker.com/r/jabref/cloudref/.
Currently, C R is an initial idea for collaborative reference management. C R is currently targeted at research groups with around ten persons. This ensures that quality standards established in a group is followed. The group, however, has to define its quality standards. For instance, there is currently no global agreement whether one should always put the abbreviation of a conference name in parentheses to the end of the conference title. The next natural step is to create a style guide for BIBT E X and to integrate a checker into C R similar to a GitHub pull request status checker [21]. When moving to a larger user base or offering C R as public SaaS offering, additional concepts for reviewing and maintaining references have to be developed and the current voting concept has to be evaluated. For instance, the number of required positive votes is a variable to evaluate.
The history of each BIBT E X entry is stored in the backend. Since there might be different views on a BIBT E X entry, C R should be able to show the history to the user. This enables him to propose another version of the entry based on historic proposals.
To provide more features for high-quality references, we plan to use the logic package of JabRef7, transpile it to JavaScript using jsweet8, and embed it in C R . We also plan to integrate C R 's functionality into JabRef. C R is currently targeted as "Multiple Instances Service" [15]. To enable it being hosted using as "Arbitrary Instances Service," we are going to work on the backend implementation and make C R a real cloud-native application [14]. This especially includes exchanging the storage layer by a PaaS one [13].
Finding related work is still a challenging task. To ease this, we aim for integrating a) the recommender system Mr. DLib [2] into C R in a similar way it has been done for JabRef [9] or b) the user profile recommendations implemented by Bibster [11].