3 For instan e, ex eeding a spe i onden e threshold in the predi tion of the laste hnique that aims at dealing with those short omings by interpreting words sis approa hes su h as latent semanti analysis (LSA) [2℄ and expli it semanti 2.1 Con ise Semanti Analysis two well known drawba ks. First, their high dimensionality and sparsity; se and do uments in a spa e of on epts. Dierently from other semanti analysier [9℄. ond, they do not apture relationships among words. CSA is a semanti analysis Standard text representation methods su h as Bag of Words (BoW) suer of The rest of this do ument is organized as follows: Se tion 2 des ribes our with our method on the eRisk 2017 dataset. Finally, Se tion 4 depi ts potential spe i point of the time. That aspe t, that ould be named as lassi ation eRisk 2017 data set and rea hed the best (lowest) reported results up to the do uments read up to the lassi ation point as standard omplete do uments. with partial information (CPI) might be addressed with a simple approa h that onsists in training with omplete do uments as usual and onsidering the partial poral variation of terms (TVT), seems to show some interesting hara teristi s use the temporal variation of terms as on ept spa e of a re ent on ise semanrated approa hes might be used.
In this arti le we propose an original idea that expli itly onsiders the sequente hnique whi h models words and do uments in a small on ept spa e whose issue has been addressed with very simple heuristi although more elabo- rules3 proposed method for the ERD problem. Se tion 3 shows the obtained results ment. This aspe t, that we will refer as the lassi ation time de ision (CTD) should be assigned to a do ument, but also de iding when to make that assignBayes algorithm to deal with partial information. ti analysis (CSA) approa h [7℄. CSA is an interesting do ument representation future works and the obtained on lusions.
In [3℄ the CPI aspe t was onsidered by analysing the robustness of the Nave tiality of data to deal with the unbalan ed data sets problem. In a nutshell, we to deal with the ERD problem. In fa t, it obtained a robust performan e on the on epts are obtained from ategory labels. CSA has obtained good results in author proling tasks [8℄ and the variant proposed in this arti le, named temLast, but not least, an ERD system needs to onsider not only whi h lass moment for and error evaluation measures. ERDE5 ERDE50 words and text fragments in a spa e of on epts that are lose (or equal) to the uments will be represented in a spa e. That spa e size is usually q-dimensional ategory labels. For instan e, if do uments in the data set are labeled with q to standard ategory labels although, as we will see later, they might represent To explain the main on epts of the CSA te hnique we rst introdu e some basi notation that will be used in the rest of this work. Let D = {hd1, y1i, . . . , hdn, yni} more elaborate aspe ts. In this ontext, we will denote as to V = {t1, . . . , tm} the vo abulary of terms of the olle tion being analysed. be a training set formed by pairs of do uments and variables that indi- n (di) (yi) is the on ept spa e. For the moment, onsider that these on epts orrespond vo abulary size (more than 10000 or 20000 elements in general). analysis (ESA) [4℄ whi h usually require huge omputing osts, CSA interprets dierent ategory labels (usually no more than 100 elements), words and do mu h smaller than standard BoW representations whi h dire tly depend on the ate the on ept the do ument is asso iated with, where yi ∈ C C = {c1, . . . , cq}
(
3 Experimental Analysis 3.1 Data Set
https://sites.google. om/site/l agnina/resear h/Tables_eRisk17.rar
an observe that this model only re overs a 45% of the depressed individuals.
In our study we onsider the two values of used in the pilot task: o o = 5 low measure (0.19). Table 2 shows similar results when a (random F1 CSA⋆-RF Bayes lassier. Those values orrespond to the setting where an instan e is probability greater or equal than 0.8 Surprisingly, the best results for (p ≥ 0.8). and error values with respe t to the previous model. (ERDE50) forest) ombination with is used to lassify the writings of the individuals. p ≥ 0.6 all the onsidered measures are obtained on the rst hunk. In this hunk, we asso iated probability is greater (or equal) than ertain threshold p tr (p ≥ tr). onsidered as depressive if the lassier assigns to the target/positive lass a In this study we evaluated 5 dierent settings for the probabilities assigned for ability of the predi ted lass. In those ases, we an sele t dierent thresholds tr Table 1 shows the results obtained with a BoW representation and a Nave their predi tions with some kind of onden e, in general, the estimated proband In ea h hunk, lassiers usually produ e (ERDE5) o = 50 (ERDE50). onsidering that an instan e (do ument) is assigned to the target lass when its as depressed ee tively had this ondition resulting in onsequen e in a very ea h lassier: and Due to spa e p = 1, p ≥ 0.9, p ≥ 0.8, p ≥ 0.7 p ≥ 0.6. onstraints, only the best results obtained with a parti ular setting are shown.6 Here, measure is also low but we an observe a deterioration in the F1 (ERDE5) 0.40 0.47 0.35 0.24 0.14 0.75 0.06 0.04 0.15