"... In PAGE 2: ... (Morita and Shinoda, 1994) use reading duration in place of the read/ignore attribute. Table1 shows the result of combining these forms with the... In PAGE 3: ... Hence this category would not contain any reference to the actual value of a rating only the fact that a rating had, or had not, occurred. The Repeated Use category in Table1 could really refer to any of the other categories of data. However, it has an appealing analogue with conventional library practice, that of date stamps in the back of a borrowed book.... In PAGE 3: ... Systems usually allow users to read a shortened or summary version of a document; bibliographic databases often have an abstract rather than the full-text of an article. At the bottom of the list in Table1 the action Associate refers to items which are closely connected to those that are examined, e.... In PAGE 3: ...hat are examined, e.g. items in the second page of hits which is never reached by the user. The action Query refers to query terms which have been used by searchers and can then be reused by subsequent searchers who use related terms (Koenig, 1990). The collection of these types of implicit data does not pose difficult technological problems: many information access tools could easily be modified to record most of the categories of data in Table1 . In addition, there is a considerable body of research in LIS on the closely related field of transaction log analysis, e.... In PAGE 4: ... The test subjects in the Morita amp; Shinoda experiment were asked to read items continuously, a very different scenario from their usual news-reading habits. A separate source of implicit data for Web users are the files of bookmarks, or hotlists, of Web pages; the Mark category from Table1 . The Siteseer system (Rucker and Polanco, 1997) uses the overlap between bookmark files to create virtual communities of interest and then recommends URLs pages from a users apos; virtual neighbours.... ..."

"... In PAGE 2: ... (Morita and Shinoda, 1994) use reading duration in place of the read/ignore attribute. Table1 shows the result of combining these forms with the... In PAGE 3: ... Hence this category would not contain any reference to the actual value of a rating only the fact that a rating had, or had not, occurred. The Repeated Use category in Table1 could really refer to any of the other categories of data. However, it has an appealing analogue with conventional library practice, that of date stamps in the back of a borrowed book.... In PAGE 3: ... Systems usually allow users to read a shortened or summary version of a document; bibliographic databases often have an abstract rather than the full-text of an article. At the bottom of the list in Table1 the action Associate refers to items which are closely connected to those that are examined, e.... In PAGE 3: ...hat are examined, e.g. items in the second page of hits which is never reached by the user. The action Query refers to query terms which have been used by searchers and can then be reused by subsequent searchers who use related terms (Koenig, 1990). The collection of these types of implicit data does not pose difficult technological problems: many information access tools could easily be modified to record most of the categories of data in Table1 . In addition, there is a considerable body of research in LIS on the closely related field of transaction log analysis, e.... In PAGE 4: ... The test subjects in the Morita amp; Shinoda experiment were asked to read items continuously, a very different scenario from their usual news-reading habits. A separate source of implicit data for Web users are the files of bookmarks, or hotlists, of Web pages; the Mark category from Table1 . The Siteseer system (Rucker and Polanco, 1997) uses the overlap between bookmark files to create virtual communities of interest and then recommends URLs pages from a users apos; virtual neighbours.... ..."

"... In PAGE 9: ...able6: Results of the Joke Test Set.................................................................................. 79 Table7 : Non-joke results .... In PAGE 89: ... Table7 : Non-joke results The non-joke texts incorrectly identified as jokes, using regular N-grams, were based on jokes number 20, 39, 49, 50 from Appendix B: Training Jokes. The text based on KKjoke39 was identified as a joke, and its punchline made sense with the found wordplay.... ..."

"... In PAGE 2: ... (Morita and Shinoda, 1994) use reading duration in place of the read/ignore attribute. Table1 shows the result of combining these forms with the... In PAGE 3: ... Hence this category would not contain any reference to the actual value of a rating only the fact that a rating had, or had not, occurred. The Repeated Use category in Table1 could really refer to any of the other categories of data. However, it has an appealing analogue with conventional library practice, that of date stamps in the back of a borrowed book.... In PAGE 3: ... Systems usually allow users to read a shortened or summary version of a document; bibliographic databases often have an abstract rather than the full-text of an article. At the bottom of the list in Table1 the action Associate refers to items which are closely connected to those that are examined, e.... In PAGE 3: ...hat are examined, e.g. items in the second page of hits which is never reached by the user. The action Query refers to query terms which have been used by searchers and can then be reused by subsequent searchers who use related terms (Koenig, 1990). The collection of these types of implicit data does not pose difficult technological problems: many information access tools could easily be modified to record most of the categories of data in Table1 . In addition, there is a considerable body of research in LIS on the closely related field of transaction log analysis, e.... In PAGE 4: ... The test subjects in the Morita amp; Shinoda experiment were asked to read items continuously, a very different scenario from their usual news-reading habits. A separate source of implicit data for Web users are the files of bookmarks, or hotlists, of Web pages; the Mark category from Table1 . The Siteseer system (Rucker and Polanco, 1997) uses the overlap between bookmark files to create virtual communities of interest and then recommends URLs pages from a users apos; virtual neighbours.... ..."

"... In PAGE 3: ...bserved intra-cluster variance, but also minimizes the unobserved intra-cluster variance, i.e., the variance of unobserved features within each cluster. Table1 summarizes the similarities and differences of our setting to that of supervised learning. The key difference is that in supervised learning, the set of features is fixed and the training instances are assumed to be randomly drawn from some distribution.... In PAGE 6: ... Unlike the standard bounds in supervised learning, this bound increases with the number of instances (m), and decreases with increasing numbers of observed features (n). This is because in our scheme the training size is not the number of instances, but rather the number of observed features (see Table1 ). However, in the next theorem we obtain an upper bound that is independent of m, and hence is tighter for large m.... In PAGE 6: ... However, for relatively large n one can use the bound in Lemma 2, which converges faster. As shown in Table1 , Theorem 3 is clearly analogous to the standard uniform convergence results in supervised learning theory (see e.g.... In PAGE 10: ... Distance-Based Clustering In this section we extend the framework and include analysis of feature generalization bounds for distance-based clustering. We apply this to analyze feature generalization of the k-means clustering algorithm (See Table1 ). The setup in this section is the same as the setup defined in Section 2 except as described below.... ..."

"... In PAGE 17: ... TAPESTRY gave two approaches for filtering: automatic, where the system evaluates what is interesting to the user, and social, where users help each other. Table1 summarizes a few paradigmatic systems developed until now or under development, emphasizing the users profile and matching techniques. Sift and Newsweeder represent two examples of automatic filtering systems.... In PAGE 18: ... Anonymous access is possible for search, browse or even retrieval, but users are always suggested to identify themselves for profile management. Explicit Implicit Tecniques Arguments Grouplens (1992) Usenet Numeric Vector Numeric Vector (reading time) Cosine measure (user profile) versus (users profiles) Collaborative filter system Sift (1994) Usenet Keywords list - Boolean (IS) versus (user profile) Filter system Newsweeder (1994) Usenet Numeric Vector Numeric Vector (user history) Cosine measure (IS) versus (user profile) Content based-filter system Fab (1994) Web Numeric Vector Numeric Vector (user history) Cosine measure (IS) and (users profiles) versus (user profile) Collaborative and content-based filter system ReferralWeb (1994) Web (user profile) versus (community profile) Collaborative and social filtering Remarks mention of a person or a document System Information Source (IS) Profile Matching Table1... In PAGE 29: ... (Morita and Shinoda, 1994) use reading duration in place of the read/ignore attribute. Table1 shows the result of combining these forms with the... In PAGE 30: ... Hence this category would not contain any reference to the actual value of a rating only the fact that a rating had, or had not, occurred. The Repeated Use category in Table1 could really refer to any of the other categories of data. However, it has an appealing analogue with conventional library practice, that of date stamps in the back of a borrowed book.... In PAGE 30: ... Systems usually allow users to read a shortened or summary version of a document; bibliographic databases often have an abstract rather than the full-text of an article. At the bottom of the list in Table1 the action Associate refers to items which are closely connected to those that are examined, e.... In PAGE 30: ...hat are examined, e.g. items in the second page of hits which is never reached by the user. The action Query refers to query terms which have been used by searchers and can then be reused by subsequent searchers who use related terms (Koenig, 1990). The collection of these types of implicit data does not pose difficult technological problems: many information access tools could easily be modified to record most of the categories of data in Table1 . In addition, there is a considerable body of research in LIS on the closely related field of transaction log analysis, e.... In PAGE 31: ... The test subjects in the Morita amp; Shinoda experiment were asked to read items continuously, a very different scenario from their usual news-reading habits. A separate source of implicit data for Web users are the files of bookmarks, or hotlists, of Web pages; the Mark category from Table1 . The Siteseer system (Rucker and Polanco, 1997) uses the overlap between bookmark files to create virtual communities of interest and then recommends URLs pages from a users apos; virtual neighbours.... In PAGE 41: ...Page 8 rating-choices.doc 21 Jul 1997 Table1 : Whose ratings are used where? Right to rate a resource Everyone can input any rating (except limitations that you cannot rate your own or your friends apos; resources) The right to input ratings is limited in some other way, to select people most proficient at providing good ratings in some way Use of ratings An average of all ratings set by everyone or by members of your peer group. Advantage: Lots of ratings available.... ..."

"... In PAGE 12: ... Water was then stored until nitrate N and phosphorus were analyzed on each sample. Results and Discussion Seasonal DM yield ( Table1 ) for the observed grass spe cies ranged from 1,660.0 to 10,765.... In PAGE 12: ...ents. The warm-season annual contribution was 76.2,62.9, Table1 . Seasonal dry matter yieldsof grass speciesas influenced by 0-, 4-, and 8-tons/acre of applied poultry litter.... In PAGE 15: ... Main plots are split for tillage comparisons of ridge-till versus conventional tillage. The cover crops are listed in Table1 . Cotton management practices are summa rized in Table 2.... In PAGE 16: ... The wheat and clover cover crops in 1994 yielded signifi cantly higher than wheat and vetch alone as shown in Table 1. Table1 also points out the combination of wheat with a legume resulted in higher yields than the legume alone. The reason for this is possibly because while the legume fixes N, the wheat acts like a scavenger and in its uptake produces more plant mass that better facilitates the following cotton crop.... In PAGE 16: ... The reason for this is possibly because while the legume fixes N, the wheat acts like a scavenger and in its uptake produces more plant mass that better facilitates the following cotton crop. The yield difference between conventional tillage and ridge-tillage was not significant, Table1 . This could be due in part to the good 1994 crop year, but more particularly to the cultivation performed on the test after emergence.... In PAGE 16: ... All other soil characters are being increased. The Table1 . Yield of lint cotton at the Delta Branch Experiment Station, 1994.... In PAGE 18: ... The study included seven rotational cropping systems composed of con tinuous soybeans (monocropped), wheat-soybeans doublecropped, and five biennial rotations of which two are single crops per year and the others doublecrop systems. The cropping sequences are shown in Table1 . Additional cultur- Terry Keisling, Professor of Agronomy, University of Arkansas, NEREC, P.... In PAGE 19: ...Table1 . Cropping sequences and seedbed preparation for 11 crop production systems from 1981 to 1984.... In PAGE 23: ... The analyses of variance (ANOVA) of data were car ried out using standard statistical procedures for randomized complete block and split-plot experimental designs. Results and Discussion The YWC was about 50% dry matter and had a very high C:N ratio ( Table1 ). Cumulative applications of YWC have totaled 806,672, and 269 Mg/ha for some treatments depend ing upon the experiment during the past 3 years (Table 2).... In PAGE 23: ... The YWC applications also resulted in significant, but sporadic decreases in populations of plant parasitic nematodes (particularlyParatrichodorus minor (Col bran) Siddiqi), depending on the experiment and year (Gal laher and McSorley, 1994a; other data not shown). Increased forage yields of field corn from use of YWC Table1 . Analysis of yard waste compost used on the Haufler and Agronomy farm research experiments in 1992, 1993, and 1994.... In PAGE 26: ... The total N applied in the completefertilizer was 96 lb/Acre and should have been adequate for high yield tobacco under Florida conditions. Leaf analysis showed that average N con centration increased by 76% from the 0 lb N/A treatment to the 75 lb N/A treatment ( Table1 ). This indicated that either not enough N was applied or that the excess rainfall/irriga tion did, in fact, leach N below the tobacco roots.... In PAGE 26: ... Nitrogen concentration in the diagnostic leaf was positively related to dry matter yield (Tables 1 amp; 2). Leaf yield respond ed to 50 lb supplemental N/A, stalk yield to between 25 Table1 . No-tillage tobacco leaf N concentration from weed con trol and supplemental N treatments.... In PAGE 30: ...Table1 . Effects of tillage and lime rate on population densities of ring nematodes (Criconemella ornata at planting and har vest of soybeans.... In PAGE 30: ...01; ns = not significant. Results and Discussion Population densities of the ring nematode were significantly lower in conventional-tillage plots than in no-tillage plots on one of two sampling dates ( Table1 ). The root-knot nema tode showed a similar response, with lower densities in conventional-tillage plots on both sampling dates (Table 2), as did the stubby-root nematode, which was significantly low er in conventional-tillage plots on one sampling date (Table 3).... In PAGE 32: ...3, 1994 (year 3). Year 3 concluded on Oct. 14 1994. Results and Discussion Winter 1991 Unusually dry fall conditions ( Table1 ) delayed soil moisture recharge, and prevented appreciable tile drainage until the end of December 1991 (Figure 1). Winter drainage was es sentially complete by the end of February 1992.... In PAGE 32: ... Winter drainage was es sentially complete by the end of February 1992. From then on, lower than normal spring rainfall ( Table1 ) and increas ing evapotranspiration prevented significant drainage for the rest of the fallow period. Cumulative drainage was consis tently less under rye than it was under fallow (Figure 1).... In PAGE 33: ...47 tons/A when it was killed, and it contained 84 lb N/A . Y Summer 1992 Oct Nov Dec Jan Feb Mar Apr Despite below-normal rainfall for March, April, and May, above average amounts from June through September 1992 Table1 . Monthly rainfall from October 1991 through October 1994, and long-term (1884-1991) average monthly rainfall at Watkinsville, GA.... In PAGE 33: ...28 -0.55 ( Table1 ) generated considerable summer drainage (Figure 2). During summer 1992, cumulative drainage was greater November 7.... In PAGE 35: ...or CT; P lt;0.02). 45 40 30 35 25 20 $0 8 15: 10 5- Summer 1993 - Leachaie NO,-N amp; concentratio Summer 1993 apos; - -. - Not measurable - This corn growing season was one of the driest on record ( Table1 , Figure 4). Only traces of drainage occurred, which provided insufficient sample volumes for N03-N analysis.... In PAGE 36: ...or NT, 63.0 lb/A for CT; P lt;0.12) was affected significant ly by tillage. Summer 1994 In contrast to summer 1993, this corn growing season was much above average in rainfall ( Table1 ). Total drainage dur ing summer 1994 was significantly affected by tillage (8.... In PAGE 38: ...preplant applicationwas necessary because it took that long for the atrazine to be leached into the root zone and kill the tall fescue. At least 3 qt/A of atrazine 4L were required to kill the tall fescue ( Table1 ). The addition of Gramoxone Ex tra allowed the grower to use less atrazine.... In PAGE 38: ... The addition of Gramoxone Ex tra allowed the grower to use less atrazine. Results of UK Table1 . Tall fescue control 8 weeks after treatment (WAT) with atrazine and atrazine combinations in 1967.... In PAGE 40: ... In ad dition, as Kruger had indicated, the no-tillage sample shows lower bulk densities than does the conventional sample. Table1 shows the land use and management of sites sam pled, the soil series, the maximum bulk density, and the per- 2.00 1.... In PAGE 41: ...056 and 0.077 g cm-3 per Table1 . Soil, county, land use, maximum bulk density, and percent organic carbon in soils used in the experiments.... In PAGE 44: ... Joseph, grain yield did not differ between tillage treat ments (Table 2). As a result of delayed seedbed preparation due to wet soil conditions, the CT treatment was planted ap- Table1 . Influence of tillage, cover crop, and N rate on corn grain yield, plants per acre (PPA), ears per acre (EPA), kernel weight, and kernels per ear at St.... In PAGE 52: ... However, canopy cover was adequate by mid-April (data not presented). In 1993, canopy cover of na tive weeds was significantly higher in March and April than cover from rye or the legumes ( Table1 ). Rye and wheat in- creased canopy cover during February and March of 1994 more than the other species.... In PAGE 52: ... No unusual problems occurred at planting or during the potato growing seasons except in 1994. Soil moisture levels Table1 . Cover crop canopy cover, by dates, 1993-1994.... In PAGE 55: ...t the 0.05 probability level. Results and Discussion The first year (1992) was the establishment year; therefore, the data being reported for both locations are for 1993 and 1994. Rainfall for the 1993-1994 growing season of May- October ranged from 19 to 40 inches ( Table1 ). Rainfall for 1993growing season ranged from normal for Prairie to above normal for Verona.... In PAGE 55: ... The lack of yield difference and the lower yield may have been due to stunting from post emergenceherbicide in- jury.On the Leeper soil for 1993, there was no corn yield differ- Table1 . 1993-1994 rainfall at Prairie Research Unit, Prairie, MS and Northeast Mississippi Branch Station, Verona, MS.... In PAGE 58: ... Only the tilled plots in 1992 produced enough ryegrass in the fall to utilize. The December 1992 ryegrass yield from tilled plots was significantly higher than for either herbicide or live sod plots ( Table1 ). The 50, 100, and 150 lb/A N rates applied to tilled plots in fall 1992 had high enough forage yields to have been grazed, whereas neither sod at any N rate had sufficient fall forage growth either year (Tables I and 2).... In PAGE 59: ...Table1 . Ryegrass yield as affected by seedbed and N rate, 1992-1993.... In PAGE 61: ...01-686-5422).Robert M. Zablotowiczand LewisA.Gaston, USDA-ARS, Southern Weed Science Laboratory, Stoneville, MS. silt loam soil (fine-silty, mixed, thermic Aeric Ochraqualf) from conventional tillage and no-tillage soybean (Glycine max) plots ( Table1 ). Fluometuron dissolved in 0.... In PAGE 61: ... Fluometuron sorption kinetics were described using a three-site, reversible model (Gaston and Locke, 1994), and sorption at selected shaking times (1, 24, and 96 hours) was evaluated using the Freundlich equation (xm = Nonlinear regression was used to calculate Kf and n-1coeffi- cients in the Freundlich equation. Table1 . Characteristicsof Dundee silt loam (0 - 5 cm) in con ventional tillage (CT) and no-tillage (NT).... In PAGE 66: ...t the 0.05 probability level. Discussion Corn Prairie. Winter VCC development was slow from October 1992 and 1993, through March 1993 and 1994, on all treat ments ( Table1 ). Maximum VCC was 20% for RT2 soybeans followed by RT1 corn on March 8, 1993 and 42% VCC on Feb.... In PAGE 66: ... Stale seedbed (NT, RTl) continuous corn and corn rota tion stale seedbed treatment RT2 soybeans followed byRTI corn had similar GRC. Both treatments had 53 and 76% or more GRC after planting corn April 14, 1993 and April 20, 1994, Table1 . Effect of corn rotation and tillage on winter vegetation canopy cover and ground residue cover on a Vaiden silty clay soil from October 1992 through May 1994 at the MAFES Prairie Unit, Prairie, MS.... In PAGE 70: ... When previous erosion was a fac tor, the studies conducted on sites that were moderate to sever ly eroded had a higher reduction in NT yields, 32 %, when compared to CT than the noneroded or slightly eroded sites, 24% when compared to CT. Table1 . Summary of yield performance of 21 soybean tillage trials conducted at the NMBES from 1978 to 1987 where the area was in CT soybean production the previous year before the trial.... In PAGE 73: ... The 5% level of significance was used. Results General Yields were taken for the winter wheat crop of 1993-1994 and the soybean crop of 1994 ( Table1 ). Soil cone indices shown below were from the spring soybean and fall wheat crops of 1994 (Table 2).... In PAGE 73: ...4 Mean * Soybeans or wheat, surface or deep-tilledmeans with the same letter are not significantly different using the LSD separation procedure. Yield (Fall 1993 and Spring 1994) For the wheat planted in 1993, yields were 3 bu/A higher for disked than for nondisked treatments ( Table1 ). This was probably a result of poorer stand in nondisked plots.... In PAGE 73: ...ore (9.9 bu/A) than no deep-tillage treatments. For the 1994soybean, yields were 14.9bu/a higher for non disked than for disked treatments ( Table1 ). This could be at least partly a result of the 1.... In PAGE 76: ...Table1 . Selected site characteristics, cultural practices, and temporal log for tillage experi ments at NEREC, Keiser; CBES, Marianna; and PTES, Colt.... In PAGE 78: ..., 1987; Hubbard and Sheridan, 1989). Multiple cropping forage system possibilities are numer ous for dairy producers ( Table1 ) (Gallaher and Cummings, 1976; Gallaher, et al., 1991; Johnson, et al.... In PAGE 78: ... of applying fertilizers, net return would increase to $536/acre, a net $214/acre increase. Choice of a dairy forage production system that produces high quality forage ( Table1 ) and best utilizes recycled nutrient wastewater would be dependent upon nutrient concentrations and ratios in the dairy wastewater. Nitrogen losses from soil in the form of ammonia (NH3) can range from 20 to 90% of applied N (Hargrove, 1988).... In PAGE 79: ...Table1 . Nitrogen and P removal by multiple cropping systems.... In PAGE 84: ... Results and Discussion Manure application rates were not consistent from one ap plication date to the next (Table I). The December 1992 (fall) application was somewhat under the target rate of 5 tons dry Table1 . Manure applications made during the study period.... In PAGE 85: ... Apparent losses of nitrate from the soil profile between November 1993 and April 19994were between 1 and 29 lb N/A in experiment I (Table 3). In experiment11, plots receiv ing fall manure in November, 1993( Table1 ) evidenced gains of 12-14 lb N/A in profile nitrate over this period. In ex periment I, fertilizer N treatments were less apparent in these data than in those of April 1993, but spring manure applica tions were more evident (Table 3).... In PAGE 89: ... Results will be discussed by year since varieties were not con sistent with respect to year. Results and Discussion Tillage had no influence on days to 50% heading in 1993 Table1 . Influence of tillage and N rate on performance and grain yield of drill-seeded rice varieties.... In PAGE 90: ...Table1 ). Higher N resulted in a modest increase in days to 50%heading.... In PAGE 93: ...Zablotowicz et al., 1995). Microbial biomass N was deter- mined by chloroform fumigation,K2SO4extraction, and nin hydrin reaction as described by Joergensen and Brookes (1990). Results and Discussion In the cotton study, annual ryegrass residues in both CT and NT significantly enhanced all measured microbial popu lations ( Table1 ). Stimulation of soil microorganisms was more persistent in the surface soil (0-2 cm) in NT- ryegrass, while greater microbial populations were observed in the 2-10 cm depth of CT-ryegrass plots.... In PAGE 93: ... Long-term NT soils croppedwith corn in several locations acrossthe United States had populations of aerobic bacteria, facultative anaerobes, nitrite oxidizers, and fungi that were higher than those found in soils under CT (Doran, 1980). Only slight increases in microbial populations were observed after 4 years of NT in Table1 . Effect of tillage and annual ryegrass cover crop on microbial populations of a Dundee silt loam, cotton study, 1994.... In PAGE 99: ...ey et al., 1993). After the fifth crop year, the study was revised. No-tillage replaced minimum and ridge tillage, corn replaced grain sor ghum, and full-season no-tillage soybeans and a corn-cotton rotation were initiated ( Table1 ). The first-phase conventional and no-tillage were retained.... In PAGE 99: ..., 1993). Table1 . Tillage and cropping treatments for phase two of the tillage and rotation study.... In PAGE 107: ...=0.05 level. Results and Discussion As stated previously, drought resulted in no cotton yield in the experiment in 1993. Total N applied in that year was 80 lb N/acre and the 1993-1994 rye winter cover produced abundant biomass because of the high amounts of residual Table1 . Winter cover biomass production.... In PAGE 110: ...5 thrips per seedling. Insecticide appli- Table1 . Percent composition of two lepidopterous species in crimson clover during May and cotton during June, July, and August 1993.... In PAGE 115: ...Table1 . Preemergence treatments on no-till cotton.... In PAGE 121: ...Table1 . Main effects on cover crop persistence, 1994.... ..."

"... In PAGE 6: ...hat can nd higher order relationships in the feature space i.e. the dictionary words. One of the main distinguishing characteristics of our approach is that we will use a rule induction model for our representation. An example of these type of rules is illustrated in Table2 . This example is from one of our experiments using universal dictionaries comprising of single words and phrases.... In PAGE 10: ... A result of applying Swap-1 to a training set of cases results in a set of rules, and the associated error rates on the training as well as test samples. The results for applying the rule set of Table2 are illustrated in Table 5. More in-depth discussions of the Swap-1 algorithm appear in [ Weiss and Indurkhya, 1993 ] .... ..."

"... In PAGE 14: ... The data were then divided into a training set containing two- thirds of the samples and a test set containing the remaining third. All the reported results except those reported in Table1 are computed for the test set. The functional classification was obtained from the Munich Information Center for Protein Sequences Yeast Genome Database (MYGD).... In PAGE 16: ... To characterize all the genes, the learning and the test sets were now combined. In Table1 , each gene is assigned to the cluster having the largest value of the membership function for that gene. The table reveals that many subclasses are concentrated in one of the clusters found by our algorithm.... In PAGE 16: ...s 1.2 bits for our approach and 0.92 for the other two. In Table1 , produced by our method, three of the subclasses (c, e, and f) have been clearly divided into two clusters, suggesting a possible biologi- cally interesting division. Its relevance will be determined later by further biological inspection; in this article, our goal is to demonstrate that the semisupervised clustering approach can be used to explore the data set and... ..."

"... In PAGE 14: ... The data were then divided into a training set containing two- thirds of the samples and a test set containing the remaining third. All the reported results except those reported in Table1 are computed for the test set. The functional classification was obtained from the Munich Information Center for Protein Sequences Yeast Genome Database (MYGD).... In PAGE 16: ... To characterize all the genes, the learning and the test sets were now combined. In Table1 , each gene is assigned to the cluster having the largest value of the membership function for that gene. The table reveals that many subclasses are concentrated in one of the clusters found by our algorithm.... In PAGE 16: ...s 1.2 bits for our approach and 0.92 for the other two. In Table1 , produced by our method, three of the subclasses (c, e, and f) have been clearly divided into two clusters, suggesting a possible biologi- cally interesting division. Its relevance will be determined later by further biological inspection; in this article, our goal is to demonstrate that the semisupervised clustering approach can be used to explore the data set and... ..."