2 Historical introduction

The ADS Abstract Service had its beginnings at the conference Astronomy from Large Data-bases, held in Garching in 1987. There discussed the desirability of building a natural language interface to a set of astronomical abstracts (Astronomy and Astrophysics Abstracts (A&AA) was the model) using software from Information Access Systems, Inc. (IAS; E. Busch was the president of IAS). discussed the existing abstract services. At this meeting G. Shaw (who was representing IAS) saw the paper by , and noticed that the vector space classification methods developed by M.J. Kurtz for the numerical classification of stellar spectra were very similar to those developed by P.G. for the classification (and thus natural language indexing) of text. Ossorio's methods were the basis of the proposal by ; Ossorio was the founder of IAS. Shaw suggested Ossorio and Kurtz meet. Also at this conference presented the NASA plan for an astrophysics data system, and Shaw met G. Squibb.

This meeting of Kurtz and Ossorio took place in January 1988, in Boulder, CO. By the end of the meeting it was clear that the technical difficulties involved in creating an abstract service with a natural language index could be overcome, if the data could become available. A preliminary mathematical solution to the problem was developed, under the assumption that A&AA would be the source of the abstracts. This technique was later called the "statistical Factor Space,'' factor analysis being one of the tools used to create the vector space.

Over the next year NASA moved to implement the plan for the establishment of a network based, distributed system for access and management of NASA astrophysics data holdings, the Astrophysics Data System. Shaw and Ossorio founded a new company, Ellery Systems, Inc., which obtained the systems integration contract for the ADS. During this time Shaw, Ossorio, Kurtz, and S.S. Murray all spoke often about the abstract service as an integral part of the emerging ADS system, and the abstract service, and Factor Space, became nearly synonymous with the ADS project. No actual work was done to implement the abstract service during this time, Ossorio and Kurtz worked on applying their vector space classification techniques to galaxy morphologies (Ossorio & Kurtz 1989; Kurtz et al. 1990), while Murray used the original, non-statistical, Factor Space methods of to build a small ($\sim$40 documents) natural language indexing system for demonstration purposes.

During the next three years the ADS was built (Good 1992), but without a literature retrieval service, which was listed as a future development. No NASA funds were devoted to the abstract service during this time. Independently Kurtz and Watson set out to obtain the data necessary to build a prototype system; keyword data was received from the IAU (International Astronomical Union) Thesaurus project (Shobbrook & Shobbrook 1992; Shobbrook & Shobbrook 1993), and from the NASA Scientific and Technical Information (STI) branch (Pinnelli 1990). A breakthrough occurred in mid 1990 when the Astronomische Rechen Institut graciously provided Watson with magnetic tape copies of the two 1989 volumes of Astronomy and Astrophysics Abstracts. By the end of 1990 Kurtz (1991, 1992) had built a prototype abstract retrieval system, based on the statistical Factor Space.

In April, 1991 F. Giovane and C. Pilachowski organized a meeting near Washington, DC on "On-Line Literature in Astronomy.'' At this meeting discussed the desire of the American Astronomical Society (AAS) to publish on-line journals, Kurtz (1991) discussed the prototype system, and pointed out the types of queries which would be made possible if a natural language abstract system were combined with the Strasbourg Data Centers's (CDS) SIMBAD (Egret & Wenger 1988) database and with the Institute for Scientific Information's Science Citation Index (Garfield 1979), and discussed the desire of the National Space Science Data Center (NSSDC) to create a database of scanned bitmaps of journal articles. Also at this meeting were representatives of NASA's STI branch, who indicated that they would be willing to provide the abstracts from the STI (often called NASA RECON) abstracts database (Wente 1990).

Near the end of the meeting outlined the possibilities inherent in the previous talks. He described a networked data system where a natural language query system for the STI abstracts would work jointly with the CDS/SIMBAD object name index to point astronomers to relevant abstracts, article bitmaps, and electronic journal articles. Save that the World Wide Web (Berners-Lee 1994) has taken the place of the proprietary network software created for the ADS project by Ellery Systems Inc., and that the ADS has taken over responsibility for the bitmaps from the NSSDC, the current system is essentially identical to the one predicted by .

Following the meeting the NSSDC group (Warnock et al. 1993) organized the STELAR project, which held a series of meetings where many of the issues involved in electronic journals were discussed, and a consensus was reached on allowed uses of copyrighted journal article bitmaps.

In the spring of 1992 Murray took over the direct management of the ADS project; G. Eichhorn was hired as project manager. The decision was made to proceed forthwith with the development of an abstract service based on the STI abstracts. Because the STI abstract system is differently structured than the A&AA system the statistical Factor Space was abandoned in favor of a more traditional entropy matching technique (Salton & McGill 1983, see SEARCH).

The new system was working with a static database by fall, and was shown at the Astronomical Data Analysis Software and Systems II meeting in Boston (Kurtz et al. 1993). The production system was released in February 1993, as part of the package of ADS services, still part of the proprietary ADS network system. Abstract Service use quickly became more than half of all ADS use.

By summer 1993 a connection had been made between the ADS and SIMBAD, permitting users to combine natural language subject matter queries with astronomical object name queries (Grant et al. 1994). This connection was enabled by the use of the bibcode (see DATA). We believe this is the first time an internet connection was made to permit the routine, simultaneous, real-time interrogation of transatlanticly separated scientific databases.

By early 1994 The World Wide Web (Bernerz-Lee 1994) had matured to where it was possible to make the ADS Abstract Service available via a web forms interface; this was released in February. Within five weeks of the initial WWW release use of the Abstract Service quadrupled (from 400 to 1600 users per month).

By the end of 1994 the ADS project had again been restructured, leaving primarily the WWW based Abstract Service as its principal service. Also the STELAR project at NSSDC ended, and the ADS took over responsibility for creating the database of bitmaps.

The first full article bitmaps, which were of Astrophysical Journal Letters articles, were put on-line in December 1994 (Eichhorn et al. 1994). By the summer of 1995 the bitmaps were current and complete going back ten years. At that time the Electronic ApJ Letters (Boyce 1995) went on-line. From the start the ADS indexed the EApJL, and pointed to the electronic version. Also from the beginning the reference section of the EApJL pointed (via WWW hyperlinks) to the ADS abstracts for articles referenced in the articles; again this was enabled by the use of the bibcode.

Also during this time the NASA STI branch became unable to provide abstracts of the journal articles in astronomy. In order to continue the abstract service cooperative arrangements were made with nearly every astronomical research journal, as well as a number of other sources of bibliographic information. DATA describes these arrangements in detail.

The next year (1996) saw nearly every astronomy journal which had not already joined into collaboration with ADS join. Also in 1996 the American Astronomical Society purchased the right to use a subset of the Science Citation Index, and gave these data to ADS (Kurtz et al. 1996).