Une étude comparative de six annuaires du Web francophone

Thomas

Informations

• Publié par: Phewyung
• Nombre de lectures: 15
• Langue: English

Extrait

Houssem Assadi and Thomas Beauvisage

France Télécom R&D – DIH/UCE, France

A Comparative Study of Six French-language Web

Directories

Abstract:

This paper presents a comparative study of six French-language Web directories (MSN,

Nomade, Open Directory, Voila, Voila Pages Perso and Yahoo). The study focuses on the quantitative

and qualitative aspects of the organization of these directories, and on the way in which they describe

sites. It reveals a wide variety of structures, content and organizational principles. In this respect, Web

directories do not correspond to classic theories of classification. They highlight the difficulty of

proposing a structured representation of the heterogeneous content of the Web.

1. Introduction

Web directories, lists of web sites classified into categories, are extensively used by

Internauts. They are particularly useful for novices since they facilitate navigation and provide

particularly relevant links, especially compared to "full text" search engines with their

problems of complexity of queries, noise, etc. The creation of Web directory categories and

the classification of sites are done manually, which contributes to an acceptable organization

of information: categories are created "rationally" and systematically, and the relevance of

listed sites is generally guaranteed.

We undertook a comparative study of six Web directories partially or totally dedicated

to the French-language Web:

MSN

,

Nomade

,

Open Directory

,

Voila

,

Voila Pages Perso

and

Yahoo

. This study concerned data available in these Web directories in February 2001. We

first developed a specific software package in order to explore the structure and content of

directories (hierarchical links and cross-references between categories, listed addresses and

site descriptions). On the basis of these data we then performed a qualitative analysis of the

organizational principles of each directory, followed by quantitative investigations consisting

of: 1) calculation of statistical indicators representing the structure and complexity of each

directory; 2) calculation of the specific characteristics of each directory, based on the content

and presentation of the indexed sites. Our study included other analyses based on a formal

exploitation of directories as graphs, but the limited scope of this article does not enable us to

include those aspects as well.

2. What is a Web directory?

A Web directory offers users lists of sites grouped into hierarchically structured

categories forming a "tree". These categories contain the indexed sites or pages (identified by

their URL, i.e. their Internet address), along with a brief description of their content.

Directories differ not only in size and number of URLs presented and indexed, but also

in structure. The structure of a directory can be defined by the combination of three elements:

-

Multi-indexing: some directories index the same URL in different categories; thus, the

same address sometimes appears several times in different places in a directory.

Position of the URLs indexed in the tree: some directories propose URLs in all their

categories, while others classify them only in terminal categories (i.e. those which do not

have sub-categories).

-

Use of cross-references: directories such as Yahoo propose links, within categories, to

other categories which may be situated anywhere in the directory and not necessarily

directly beneath them in the tree.

Structurally, each directory is thus a combination of these three elements, as presented

in Table 1.

Multi-indexing

URLs are indexed only

in terminal categories

Use of cross-references

MSN

Nomade

Open Directory

Voila

Voila PP

Yahoo

Table 1 - Structural description of directories

3. Differences of size and structure

The first difference we observe is in the

coverage

of directories, calculated in terms of

the number of indexed URLs. This calculation must take into account multi-indexing: if a

directory lists the same URL in several places, it will present more addresses than individual

URLs actually indexed. That is why it is important to distinguish between the number of

URLs presented and the number of individual URLs indexed (cf. Table 2). For example,

whereas Nomade is the directory with the broadest coverage (139,000 indexed URLs), Yahoo

presents the highest number of URLs to the user and thus has the highest repetition rate.

Directories also vary highly in terms of

depth

. Yahoo, with its 18-level depth tree, has

the most depth. The smallest directory, Voila Pages Perso, has a maximum depth of only 5.

But depth does not systematically correlate with the number of URLs (see Table 2). We

furthermore found wide variability in terms of

density

(mean number of URLs per category).

Whereas Nomade and Voila propose a mean of close to 20 URLs per category containing at

least one URL, Open Directory, Yahoo and MSN offer a mean of between 5 and 10.

Number of

categories

Maximum

depth

Total

number of

URLs

presented

Number of

individual

URLs

Mean rate of

URL

repetition

Number of

categories

which index

at least one

URL

Mean

number of

URLs per

category

with at least

one URL

MSN

6,875

7

62,523

46,137

1.35

6,507

9.61

Nomade

9,165

10

183,590

138,832

1.32

8,754

20.97

Open Dir.

5,244

10

32,496

32,496

1

4,201

7.73

Voila

8,967

10

134,502

59,744

2.25

7,854

17.12

Voila PP

601

5

50,794

27,923

1.81

564

90.06

Yahoo

44,137

18

192,160

106,832

1.8

37,178

5.17

Table 2 - Indicators of size, depth and multi-indexing

Finally, cross-references strongly influence the directory structure. They facilitate

navigation for the user and enable the creators of directories to compensate for the rigidity of

the overall organization of the tree. The four directories that use cross-references (Nomade,

Open Directory, Voila and Yahoo) do so differently (see Table 3). While Nomade and Voila

make little use of them (only 1.6% of Voila categories use cross-references, with a mean of

1.4 cross-references proposed), Open Directory and Yahoo use them on a big scale (20% of

the Yahoo categories, with a mean of 4 cross-references per category).

Total number of

categories

Number of

categories with a

cross-reference

Total number of

cross-references

Mean number of

cross-references

per category with

cross-reference

Proportion of

categories with

cross-reference

MSN

6,875

-

-

-

-

Nomade

9,165

586

997

1.70

6.3%

Open Dir.

5,244

526

1,660

3.15

10.0%

Voila

8,967

142

202

1.42

1.6%

Voila PP

601

-

-

-

-

Yahoo

44,137

8,806

34,840

3.95

19.9%

Table 3 - Use of cross-references

4. Web directories have different organizational principles

In this section we consider the principles governing the organization and structuring of

directories. Several models can be used for the organization of data and knowledge, derived

from fields as varied as knowledge representation in Artificial Intelligence, the compilation of

thesauruses (e.g. in libraries) or the creation of indexes and other "paper" directories (e.g.

yellow pages, professional directories). These different models have been adopted by the

publishers of Web directories, in some cases intentionally, in others less so. Three modes of

organization can be distinguished:

-

Systematic categorization of domains of human activities, objects of daily life, etc. in

an ontological approach.

-

Less systematic and more practical cataloguing, focused on human activities (e.g.

business, recreation, various forms of sociability), in a "yellow pages" approach.

-

Categorization of the "Internet world": mapping of sites and services available on the

Internet, without any precise criteria for classification of objects of the world, human

activities, etc.

Note that Internet directories, at least those that we studied, do not correspond strictly to

any one of these approaches; they combine the principles of usual classificatory objects:

ontologies, thesaurus, etc. In the six directories that we studied, very different organizational

models were identified.

For example, if we compare Yahoo and Voila, Yahoo has a systematic classification

approach evident in a large number of categories (44,000 compared to 9,000 in Voila)

organized in an 18-level tree structure (compared to 10 levels in Voila). Yahoo also has a very

dense network based on a system of cross-references between categories (34,000 cross-

references, compared to 200 in Voila). The main first-level categories in Yahoo are

"Regional" (

Exploration géographique

) and "Business and economy" (

Commerce et

économie

). This reveals a systematic classification approach. The encyclopaedic aspect of the

Yahoo directory is also evident in the presence of categories such as "Social science"

(

Sciences humaines

) at the top of the tree. By contrast, Voila has a pragmatic approach,

focused on services related to human activities: business, social activities and recreation. This

practical aspect of Voila is strengthened by the existence of a first-level category "Shopping,

daily life" (

Achat, vie pratique

) which accounts for 13% of all indexed sites and has no

equivalent at the top level in Yahoo France directory.

5. Web directories have little in common

The six directories studied index a total of 283,000 distinct URLs. We found that in

general the directories overlap very little and have only 25 URLs in common; 70% of all the

indexed URLs are in only one directory and 89% in one or two directories.

Each directory has its specific characteristics, confirmed by a low two-to-two overlap.

Since the sizes of directories differ, the calculation of two-to-two overlaps between directories

is asymmetrical and must be analysed for each pair of directories (Table 4).

Ð

shares n % of its

URL with

Î

MSN

Nomade

Open

Directory

Voila

Voila PP

Yahoo

MSN

100.0%

33.8%

13.2%

22.7%

1.5%

35.7%

Nomade

11.2%

100.0%

9.3%

19.7%

3.4%

30.4%

Open Directory

18.8%

39.9%

100.0%

22.9%

2.4%

33.6%

Voila

17.6%

45.8%

12.4%

100.0%

3.8%

37.7%

Voila PP

2.4%

16.6%

2.7%

8.0%

100.0%

8.7%

Yahoo

1.4%

39.5%

10.2%

21.1%

2.3%

100.0%

Key: 11.2% of URLs of Nomade are also indexed by MSN, while 33.8% of the URLs of MSN are in

the Nomade base.

Table 4 - Overlaps between directories

The particularity of the Voila Pages Perso personal sites directory is strongly confirmed

by the very low level of overlap with other directories, especially

from

Voila Pages Perso

to

other directories, even though Voila Pages Perso is the smallest directory of all.

The mean overlap rate between the different directories is 18.1% and 22.8% if we

exclude the very specific Voila Pages Perso. As in the case of Voila Pages Perso, size does

not seem to be the decisive factor in overlaps between directories. For example, MSN and

Open Directory, both small, share less than a third of their URLs, on average, with other

directories up to four times bigger – an overlap equivalent to that between Nomade and

Yahoo, the two biggest. It thus appears that each directory indexes sites peculiar to it. This is

confirmed by the proportion, in each directory, of URLs indexed in no other directory (see

Table 5).

Directory

Number of sites indexed

Number of sites peculiar

to the directory

Percentage of sites

peculiar to the directory

MSN

46,223

21,174

45.8%

Nomade

139,051

74,089

53.3%

Open Directory

32,496

12,292

37.8%

Voila

59,801

21,482

35.9%

Voila PP

28,330

20,614

72.8%

Yahoo

107,052

48,899

45.7%

Table 5 - Proportion of indexed URLs peculiar to each directory

With the exception of Voila Pages Perso, which has a particular content (73% of URLs

peculiar to it), we note that Nomade, the biggest directory, is also the one with the most

particular characteristics (53.3%). This result was expected. Less predictable were the rates of

particularity of MSN (45.8% of URLs peculiar to it), which is three times smaller than

Nomade, and Yahoo, which has few particularities despite its size. It thus seems that there is a

twofold effect contributing to the particularity of directories: their size, which statistically

increases their chances of indexing sites that others exclude, and their content, determined by

their choice of indexed sites.

6. Web directories have strong identities

In this part we examined the particularity of each directory in terms of content and style.

We first wished to establish whether, on a given topic (e.g. art, sport or politics), different

directories have marked differences in choice of content and, if so, to what extent? We

excluded Voila Pages Perso from this analysis, because of its intrinsic particularity, and

retained only the five general-interest directories.

We qualified the content of the directories on the basis of the short descriptions they

give of indexed Web sites on a given theme. The following method was applied:

1. We first choose topics present at the first level, for the five directories studied;

2. We then extracted, for each directory and for all the sites classified under the chosen

topic, the directories' descriptions of those sites;

3. The corpus thus constituted was processed with a lexicometric tool, Alceste (Reinert

1993). This enabled us to identify the specific vocabulary used by each directory to

describe sites on a given topic.

The first topic selected was "Art and Humanities" (

Art et culture

) present at the top level

of the five directories. At this level MSN focuses on North American museums (specific

vocabulary: "US", "Canada", "Montreal", "New York", etc.), Open Directory gives priority to

downloading of music ("MP3", "server", "free", etc.), Nomade emphasizes the accessibility of

art ("consult", "invite", "share", etc.), Voila is more business-oriented ("bargains",

"shopping", "catalogue", "order", etc.) and Yahoo seems to be more eclectic, with a slight

preference for cinema ("scenario", "critique", "synopsis", etc.).

The same analysis was applied to the category "Business and economy" (

Commerce et

économie

), also found at the first level of the five general-interest directories. Here again,

MSN focuses largely on North American sites, with particular emphasis on financial topics

("financial", "bank", "investment", "shares", etc.). Nomade seems to be oriented more towards

the tourism industry ("visit", "restaurant", "rating", "hotel", etc.). Yahoo and Open Directory

tend to favour international trade and the manufacturing sector ("truck", "rubber",

"development", "international", etc.). Lastly, Voila's position appears to be less specific

although it has a slight tendency towards local business (place names are over-represented).

MSN

Nomade

Open Dir.

Voila

Voila PP

Yahoo

Nouns

78.5%

66.9%

73.8%

74.0%

70.6%

78.9%

Verbs

5.1%

16.4%

8.4%

9.8%

12.8%

5.6%

Adjectives

14.9%

12.3%

15.1%

13.2%

12.2%

14.2%

POS

maincat.

Adverbs

1.5%

4.4%

2.7%

3.0%

4.4%

1.3%

1 SG

1.3%

0.9%

0.9%

0.2%

8.8%

1.0%

2 SG

9.0%

0.9%

9.6%

2.3%

3.7%

26.0%

3 SG

79.7%

51.2%

60.6%

54.3%

46.8%

57.2%

1 PL

1.0%

1.3%

2.9%

0.2%

3.6%

0.9%

2 PL

2.2%

38.3%

15.5%

34.6%

30.1%

3.5%

Verbsan

pronouns

person/number

3 PL

6.7%

7.4%

10.5%

8.5%

7.0%

11.5%

Table 6 - Distribution of main POS categories and person/number for each directory

In addition to this, we performed a part-of-speech tagging of all the site descriptions for

each Web directory in order to see its stylistic specificities. We thus analyzed the distribution

of main POS categories and of verbs and pronouns person/number for each directory (see

Table 6). This analysis raised an opposition between two presentational attitudes: on the one

hand, directories as "helpful guides", with an over-representation of verbs and the 2

nd

person

plural (typically: "You will find on this site…"): Nomade, Voila, Voila Pages Perso; on the

other hand, directories as "neutral information relay", with very few verbs and a dominance of

nouns, adjectives and the 3

rd

person singular: MSN, Yahoo and Open Directory.

7. Discussion

Most earlier studies on Web directories were undertaken by researchers and specialists

in library science and documentation (Bertonèche 2001) (Chan, Xia et al. 1999) (Van der

Walt 1998) (Vizine-Goetz 1996). Other studies have focused on directories as systems of

reference classification, and have used them as a resource for automatic classification of

documents (Mladenic 1998), (Labrou and Finin 1999). Two studies in particular attracted our

attention in so far as they present a comparison of several Web directories (Bertonèche 2001),

(Van der Walt 1998). In both cases the approach is purely qualitative and the findings are

very close to our own, i.e. wide diversity, even heterogeneity, among the directories studied.

Our study has the particularity of combining a qualitative approach ("manual" analysis

of directories and critical analysis of their modes of organization), and a quantitative approach

based on statistical and formal processing: exploitation of the textual content of site

descriptions in directories and of the structure of the directory as a graph.

Specialists in cataloguing and library science recommend taking into account the

different classificatory and methodological theories developed by their disciplines, and

applying them to the cataloguing of Internet sites. Our view is that this classificatory approach

cannot be transposed as such to the Internet world, for at least two reasons. The first is that the

Web is not an encyclopedia of knowledge and is not comparable to a library. It offers content,

services and, more generally, resources of all kinds, with a very wide variety of topics and

quality. The second reason lies in the diversity of contexts in which the Internet is used, and

of user profiles and needs. This is a very different situation from readers in a library or other

documentary resource center, whose needs can be determined

a priori

and whose profiles

have been defined on the basis of a long history of practice.

We agree with library science specialists on the need for more rigor in the construction

of Web directories. We also think that general-interest Web directories, like those that

currently exist, have possibly reached their limits. The management of a large number of sites

and categories seems to pose problems manifested in a lack of coherence. Two areas appear to

be emerging for further development of Web directories suited to current trends in their use:

"regional" directories, specialized by sector of activity, geographic or cultural domain, etc.,

and systems based on collaborative evaluation and cataloguing of web sites within "interest

communities".

8. References

Bertonèche, J. (2001). “L'Internet-bibliothèque : accéder au savoir ou se l'approprier?”

SPIRALE - Revue de Recherches en Education(28): 195-214.

Chan, L. M., L. Xia, et al. (1999). Structural and multilingual approaches to subject access on

the web. 65th IFLA Council and General Conference, Bangkok, Thailand.

Labrou, Y. and T. Finin (1999). Yahoo! as an Ontology Using Yahoo! Categories to Describe

Documents. CIKM.

Mladenic, D. (1998). Turning Yahoo to Automatic Web-Page Classifier. European

Conference on Artificial Intelligence.

Reinert, M. (1993). “Les "mondes lexicaux" et leur logique.” Langage et société(66): 5-39.

Van der Walt, M. (1998). The Structure of Classification Schemes Used in Internet Search

Engines. Fifth International ISKO Conference, Lille, France, Ergon Verlag.

Vizine-Goetz, D. (1996). Using Library Classification Schemes for Internet Resources. OCLC

Internet Cataloging Project Colloquium, San Antonio (Texas).