Search (9 results, page 1 of 1)

  • × author_ss:"Egghe, L."
  • × language_ss:"e"
  • × year_i:[2000 TO 2010}
  1. Egghe, L.: ¬A universal method of information retrieval evaluation : the "missing" link M and the universal IR surface (2004) 0.01 
    0.005982068 = product of:
  0.020937236 = sum of:
    0.006214436 = product of:
      0.03107218 = sum of:
        0.03107218 = weight(_text_:retrieval in 2558) [ClassicSimilarity], result of:
          0.03107218 = score(doc=2558,freq=4.0), product of:
            0.109568894 = queryWeight, product of:
              3.024915 = idf(docFreq=5836, maxDocs=44218)
              0.03622214 = queryNorm
            0.2835858 = fieldWeight in 2558, product of:
              2.0 = tf(freq=4.0), with freq of:
                4.0 = termFreq=4.0
              3.024915 = idf(docFreq=5836, maxDocs=44218)
              0.046875 = fieldNorm(doc=2558)
      0.2 = coord(1/5)
    0.0147228 = product of:
      0.0294456 = sum of:
        0.0294456 = weight(_text_:22 in 2558) [ClassicSimilarity], result of:
          0.0294456 = score(doc=2558,freq=2.0), product of:
            0.12684377 = queryWeight, product of:
              3.5018296 = idf(docFreq=3622, maxDocs=44218)
              0.03622214 = queryNorm
            0.23214069 = fieldWeight in 2558, product of:
              1.4142135 = tf(freq=2.0), with freq of:
                2.0 = termFreq=2.0
              3.5018296 = idf(docFreq=3622, maxDocs=44218)
              0.046875 = fieldNorm(doc=2558)
      0.5 = coord(1/2)
  0.2857143 = coord(2/7)
    Abstract
    The paper shows that the present evaluation methods in information retrieval (basically recall R and precision P and in some cases fallout F ) lack universal comparability in the sense that their values depend on the generality of the IR problem. A solution is given by using all "parts" of the database, including the non-relevant documents and also the not-retrieved documents. It turns out that the solution is given by introducing the measure M being the fraction of the not-retrieved documents that are relevant (hence the "miss" measure). We prove that - independent of the IR problem or of the IR action - the quadruple (P,R,F,M) belongs to a universal IR surface, being the same for all IR-activities. This universality is then exploited by defining a new measure for evaluation in IR allowing for unbiased comparisons of all IR results. We also show that only using one, two or even three measures from the set {P,R,F,M} necessary leads to evaluation measures that are non-universal and hence not capable of comparing different IR situations.
    Date
    14. 8.2004 19:17:22
  2. Egghe, L.: Existence theorem of the quadruple (P, R, F, M) : precision, recall, fallout and miss (2007) 0.00 
    0.0039777067 = product of:
  0.027843947 = sum of:
    0.027843947 = product of:
      0.069609866 = sum of:
        0.02197135 = weight(_text_:retrieval in 2011) [ClassicSimilarity], result of:
          0.02197135 = score(doc=2011,freq=2.0), product of:
            0.109568894 = queryWeight, product of:
              3.024915 = idf(docFreq=5836, maxDocs=44218)
              0.03622214 = queryNorm
            0.20052543 = fieldWeight in 2011, product of:
              1.4142135 = tf(freq=2.0), with freq of:
                2.0 = termFreq=2.0
              3.024915 = idf(docFreq=5836, maxDocs=44218)
              0.046875 = fieldNorm(doc=2011)
        0.047638513 = weight(_text_:system in 2011) [ClassicSimilarity], result of:
          0.047638513 = score(doc=2011,freq=8.0), product of:
            0.11408355 = queryWeight, product of:
              3.1495528 = idf(docFreq=5152, maxDocs=44218)
              0.03622214 = queryNorm
            0.41757566 = fieldWeight in 2011, product of:
              2.828427 = tf(freq=8.0), with freq of:
                8.0 = termFreq=8.0
              3.1495528 = idf(docFreq=5152, maxDocs=44218)
              0.046875 = fieldNorm(doc=2011)
      0.4 = coord(2/5)
  0.14285715 = coord(1/7)
    Abstract
    In an earlier paper [Egghe, L. (2004). A universal method of information retrieval evaluation: the "missing" link M and the universal IR surface. Information Processing and Management, 40, 21-30] we showed that, given an IR system, and if P denotes precision, R recall, F fallout and M miss (re-introduced in the paper mentioned above), we have the following relationship between P, R, F and M: P/(1-P)*(1-R)/R*F/(1-F)*(1-M)/M = 1. In this paper we prove the (more difficult) converse: given any four rational numbers in the interval ]0, 1[ satisfying the above equation, then there exists an IR system such that these four numbers (in any order) are the precision, recall, fallout and miss of this IR system. As a consequence we show that any three rational numbers in ]0, 1[ represent any three measures taken from precision, recall, fallout and miss of a certain IR system. We also show that this result is also true for two numbers instead of three.
  3. Egghe, L.: ¬The measures precision, recall, fallout and miss as a function of the number of retrieved documents and their mutual interrelations (2008) 0.00 
    0.0014796278 = product of:
  0.010357394 = sum of:
    0.010357394 = product of:
      0.051786967 = sum of:
        0.051786967 = weight(_text_:retrieval in 2067) [ClassicSimilarity], result of:
          0.051786967 = score(doc=2067,freq=16.0), product of:
            0.109568894 = queryWeight, product of:
              3.024915 = idf(docFreq=5836, maxDocs=44218)
              0.03622214 = queryNorm
            0.47264296 = fieldWeight in 2067, product of:
              4.0 = tf(freq=16.0), with freq of:
                16.0 = termFreq=16.0
              3.024915 = idf(docFreq=5836, maxDocs=44218)
              0.0390625 = fieldNorm(doc=2067)
      0.2 = coord(1/5)
  0.14285715 = coord(1/7)
    Abstract
    In this paper, for the first time, we present global curves for the measures precision, recall, fallout and miss in function of the number of retrieved documents. Different curves apply for different retrieved systems, for which we give exact definitions in terms of a retrieval density function: perverse retrieval, perfect retrieval, random retrieval, normal retrieval, hereby extending results of Buckland and Gey and of Egghe in the following sense: mathematically more advanced methods yield a better insight into these curves, more types of retrieval are considered and, very importantly, the theory is developed for the "complete" set of measures: precision, recall, fallout and miss. Next we study the interrelationships between precision, recall, fallout and miss in these different types of retrieval, hereby again extending results of Buckland and Gey (incl. a correction) and of Egghe. In the case of normal retrieval we prove that precision in function of recall and recall in function of miss is a concavely decreasing relationship while recall in function of fallout is a concavely increasing relationship. We also show, by producing examples, that the relationships between fallout and precision, miss and precision and miss and fallout are not always convex or concave.
  4. Egghe, L.: Vector retrieval, fuzzy retrieval and the universal fuzzy IR surface for IR evaluation (2004) 0.00 
    0.0012685166 = product of:
  0.008879616 = sum of:
    0.008879616 = product of:
      0.044398077 = sum of:
        0.044398077 = weight(_text_:retrieval in 2531) [ClassicSimilarity], result of:
          0.044398077 = score(doc=2531,freq=6.0), product of:
            0.109568894 = queryWeight, product of:
              3.024915 = idf(docFreq=5836, maxDocs=44218)
              0.03622214 = queryNorm
            0.40520695 = fieldWeight in 2531, product of:
              2.4494898 = tf(freq=6.0), with freq of:
                6.0 = termFreq=6.0
              3.024915 = idf(docFreq=5836, maxDocs=44218)
              0.0546875 = fieldNorm(doc=2531)
      0.2 = coord(1/5)
  0.14285715 = coord(1/7)
    Abstract
    It is shown that vector information retrieval (IR) and general fuzzy IR uses two types of fuzzy set operations: the original "Zadeh min-max operations" and the so-called "probabilistic sum and algebraic product operations". The universal IR surface, valid for classical 0-1 IR (i.e. where ordinary sets are used) and used in IR evaluation, is extended to and reproved for vector IR, using the probabilistic sum and algebraic product model. We also show (by counterexample) that, using the "Zadeh min-max" fuzzy model, yields a breakdown of this IR surface.
  5. Egghe, L.: Properties of the n-overlap vector and n-overlap similarity theory (2006) 0.00 
    9.822896E-4 = product of:
  0.0068760267 = sum of:
    0.0068760267 = product of:
      0.034380134 = sum of:
        0.034380134 = weight(_text_:system in 194) [ClassicSimilarity], result of:
          0.034380134 = score(doc=194,freq=6.0), product of:
            0.11408355 = queryWeight, product of:
              3.1495528 = idf(docFreq=5152, maxDocs=44218)
              0.03622214 = queryNorm
            0.30135927 = fieldWeight in 194, product of:
              2.4494898 = tf(freq=6.0), with freq of:
                6.0 = termFreq=6.0
              3.1495528 = idf(docFreq=5152, maxDocs=44218)
              0.0390625 = fieldNorm(doc=194)
      0.2 = coord(1/5)
  0.14285715 = coord(1/7)
    Abstract
    In the first part of this article the author defines the n-overlap vector whose coordinates consist of the fraction of the objects (e.g., books, N-grams, etc.) that belong to 1, 2, , n sets (more generally: families) (e.g., libraries, databases, etc.). With the aid of the Lorenz concentration theory, a theory of n-overlap similarity is conceived together with corresponding measures, such as the generalized Jaccard index (generalizing the well-known Jaccard index in case n 5 2). Next, the distributional form of the n-overlap vector is determined assuming certain distributions of the object's and of the set (family) sizes. In this section the decreasing power law and decreasing exponential distribution is explained for the n-overlap vector. Both item (token) n-overlap and source (type) n-overlap are studied. The n-overlap properties of objects indexed by a hierarchical system (e.g., books indexed by numbers from a UDC or Dewey system or by N-grams) are presented in the final section. The author shows how the results given in the previous section can be applied as well as how the Lorenz order of the n-overlap vector is respected by an increase or a decrease of the level of refinement in the hierarchical system (e.g., the value N in N-grams).
  6. Egghe, L.; Rousseau, R.; Hooydonk, G. van: Methods for accrediting publications to authors or countries : consequences for evaluation studies (2000) 0.00 
    9.624433E-4 = product of:
  0.0067371028 = sum of:
    0.0067371028 = product of:
      0.033685513 = sum of:
        0.033685513 = weight(_text_:system in 4384) [ClassicSimilarity], result of:
          0.033685513 = score(doc=4384,freq=4.0), product of:
            0.11408355 = queryWeight, product of:
              3.1495528 = idf(docFreq=5152, maxDocs=44218)
              0.03622214 = queryNorm
            0.29527056 = fieldWeight in 4384, product of:
              2.0 = tf(freq=4.0), with freq of:
                4.0 = termFreq=4.0
              3.1495528 = idf(docFreq=5152, maxDocs=44218)
              0.046875 = fieldNorm(doc=4384)
      0.2 = coord(1/5)
  0.14285715 = coord(1/7)
    Abstract
    One aim of science evaluation studies is to determine quantitatively the contribution of different players (authors, departments, countries) to the whole system. This information is then used to study the evolution of the system, for instance to gauge the results of special national or international programs. Taking articles as our basic data, we want to determine the exact relative contribution of each coauthor or each country. These numbers are brought together to obtain country scores, or department scores, etc. It turns out, as we will show in this article, that different scoring methods can yield totally different rankings. Conseqeuntly, a ranking between countries, universities, research groups or authors, based on one particular accrediting methods does not contain an absolute truth about their relative importance
  7. Egghe, L.: ¬The power of power laws and an interpretation of Lotkaian informetric systems as self-similar fractals (2005) 0.00 
    8.0203614E-4 = product of:
  0.0056142528 = sum of:
    0.0056142528 = product of:
      0.028071264 = sum of:
        0.028071264 = weight(_text_:system in 3466) [ClassicSimilarity], result of:
          0.028071264 = score(doc=3466,freq=4.0), product of:
            0.11408355 = queryWeight, product of:
              3.1495528 = idf(docFreq=5152, maxDocs=44218)
              0.03622214 = queryNorm
            0.24605882 = fieldWeight in 3466, product of:
              2.0 = tf(freq=4.0), with freq of:
                4.0 = termFreq=4.0
              3.1495528 = idf(docFreq=5152, maxDocs=44218)
              0.0390625 = fieldNorm(doc=3466)
      0.2 = coord(1/5)
  0.14285715 = coord(1/7)
    Abstract
    Power laws as defined in 1926 by A. Lotka are increasing in importance because they have been found valid in varied social networks including the Internet. In this article some unique properties of power laws are proven. They are shown to characterize functions with the scalefree property (also called seif-similarity property) as weIl as functions with the product property. Power laws have other desirable properties that are not shared by exponential laws, as we indicate in this paper. Specifically, Naranan (1970) proves the validity of Lotka's law based on the exponential growth of articles in journals and of the number of journals. His argument is reproduced here and a discrete-time argument is also given, yielding the same law as that of Lotka. This argument makes it possible to interpret the information production process as a seif-similar fractal and show the relation between Lotka's exponent and the (seif-similar) fractal dimension of the system. Lotkaian informetric systems are seif-similar fractals, a fact revealed by Mandelbrot (1977) in relation to nature, but is also true for random texts, which exemplify a very special type of informetric system.
  8. Egghe, L.: Type/Token-Taken informetrics (2003) 0.00 
    5.2312744E-4 = product of:
  0.003661892 = sum of:
    0.003661892 = product of:
      0.01830946 = sum of:
        0.01830946 = weight(_text_:retrieval in 1608) [ClassicSimilarity], result of:
          0.01830946 = score(doc=1608,freq=2.0), product of:
            0.109568894 = queryWeight, product of:
              3.024915 = idf(docFreq=5836, maxDocs=44218)
              0.03622214 = queryNorm
            0.16710453 = fieldWeight in 1608, product of:
              1.4142135 = tf(freq=2.0), with freq of:
                2.0 = termFreq=2.0
              3.024915 = idf(docFreq=5836, maxDocs=44218)
              0.0390625 = fieldNorm(doc=1608)
      0.2 = coord(1/5)
  0.14285715 = coord(1/7)
    Abstract
    Type/Token-Taken informetrics is a new part of informetrics that studies the use of items rather than the items itself. Here, items are the objects that are produced by the sources (e.g., journals producing articles, authors producing papers, etc.). In linguistics a source is also called a type (e.g., a word), and an item a token (e.g., the use of words in texts). In informetrics, types that occur often, for example, in a database will also be requested often, for example, in information retrieval. The relative use of these occurrences will be higher than their relative occurrences itself; hence, the name Type/ Token-Taken informetrics. This article studies the frequency distribution of Type/Token-Taken informetrics, starting from the one of Type/Token informetrics (i.e., source-item relationships). We are also studying the average number my* of item uses in Type/Token-Taken informetrics and compare this with the classical average number my in Type/Token informetrics. We show that my* >= my always, and that my* is an increasing function of my. A method is presented to actually calculate my* from my, and a given a, which is the exponent in Lotka's frequency distribution of Type/Token informetrics. We leave open the problem of developing non-Lotkaian Type/TokenTaken informetrics.
  9. Egghe, L.: Untangling Herdan's law and Heaps' law : mathematical and informetric arguments (2007) 0.00 
    5.2312744E-4 = product of:
  0.003661892 = sum of:
    0.003661892 = product of:
      0.01830946 = sum of:
        0.01830946 = weight(_text_:retrieval in 271) [ClassicSimilarity], result of:
          0.01830946 = score(doc=271,freq=2.0), product of:
            0.109568894 = queryWeight, product of:
              3.024915 = idf(docFreq=5836, maxDocs=44218)
              0.03622214 = queryNorm
            0.16710453 = fieldWeight in 271, product of:
              1.4142135 = tf(freq=2.0), with freq of:
                2.0 = termFreq=2.0
              3.024915 = idf(docFreq=5836, maxDocs=44218)
              0.0390625 = fieldNorm(doc=271)
      0.2 = coord(1/5)
  0.14285715 = coord(1/7)
    Abstract
    Herdan's law in linguistics and Heaps' law in information retrieval are different formulations of the same phenomenon. Stated briefly and in linguistic terms they state that vocabularies' sizes are concave increasing power laws of texts' sizes. This study investigates these laws from a purely mathematical and informetric point of view. A general informetric argument shows that the problem of proving these laws is, in fact, ill-posed. Using the more general terminology of sources and items, the author shows by presenting exact formulas from Lotkaian informetrics that the total number T of sources is not only a function of the total number A of items, but is also a function of several parameters (e.g., the parameters occurring in Lotka's law). Consequently, it is shown that a fixed T(or A) value can lead to different possible A (respectively, T) values. Limiting the T(A)-variability to increasing samples (e.g., in a text as done in linguistics) the author then shows, in a purely mathematical way, that for large sample sizes T~ A**phi, where phi is a constant, phi < 1 but close to 1, hence roughly, Heaps' or Herdan's law can be proved without using any linguistic or informetric argument. The author also shows that for smaller samples, a is not a constant but essentially decreases as confirmed by practical examples. Finally, an exact informetric argument on random sampling in the items shows that, in most cases, T= T(A) is a concavely increasing function, in accordance with practical examples.

Authors