本文選題：電信投訴預(yù)測(cè)　切入點(diǎn)：深度學(xué)習(xí)　出處：《計(jì)算機(jī)應(yīng)用研究》2017年05期

【摘要】：用戶投訴預(yù)測(cè)模型能有效地降低電信用戶投訴率,對(duì)企業(yè)提高用戶滿意度和競(jìng)爭(zhēng)力有著至關(guān)重要的作用。在模型訓(xùn)練過(guò)程中,由于人工設(shè)計(jì)特征的缺陷和設(shè)計(jì)過(guò)程中存在難以預(yù)估的復(fù)雜性,使得模型預(yù)測(cè)的精度和設(shè)計(jì)特征的效率不能有很大的提升。針對(duì)上述問(wèn)題,提出了一種基于深度學(xué)習(xí)的用戶投訴預(yù)測(cè)模型。該模型通過(guò)深層網(wǎng)絡(luò)特征學(xué)習(xí)單元能從電信用戶原始數(shù)據(jù)中自動(dòng)學(xué)習(xí)到適合分類(lèi)器分類(lèi)的高層非線性組合特征,并將這些高層特征輸入到傳統(tǒng)分類(lèi)器中來(lái)提高模型的精度。通過(guò)實(shí)驗(yàn)結(jié)果分析,預(yù)測(cè)模型在AUC指標(biāo)上比以往用戶投訴模型提升了7.1%,證明了該模型自動(dòng)學(xué)習(xí)特征的有效性和深度學(xué)習(xí)在電信大數(shù)據(jù)領(lǐng)域的可用性。
[Abstract]:The customer complaint prediction model can effectively reduce the rate of telecom users' complaints and play an important role in improving the customer satisfaction and competitiveness of enterprises. Because of the defects of the artificial design features and the complexity of the design process, the precision of the model prediction and the efficiency of the design features can not be greatly improved. In this paper, a user complaint prediction model based on deep learning is proposed, which can automatically learn high-level nonlinear composite features suitable for classifier classification from the original data of telecom users through the deep network feature learning unit. These high-level features are input into the traditional classifier to improve the accuracy of the model. Compared with the previous user complaint model, the predictive model improves the AUC index by 7.1, which proves the validity of the model's automatic learning feature and the usability of the in-depth learning in the field of telecom big data.
【作者單位】： 蘇州大學(xué)計(jì)算機(jī)科學(xué)與技術(shù)學(xué)院;香港城市大學(xué)創(chuàng)意媒體學(xué)院;
【基金】：國(guó)家自然科學(xué)基金資助項(xiàng)目(61373092,61033013,61272449,61202029) 江蘇省教育廳重大項(xiàng)目(12KJA520004) 江蘇省科技支撐計(jì)劃重點(diǎn)項(xiàng)目(BE2014005)
【分類(lèi)號(hào)】：F274;F626;TP181
，

本文編號(hào)：1671388