Generalized linear models with nonlinear feature transformations are widely used for large-scale regression and classification problems with sparse inputs. Memorization of feature interactions through a wide set of cross-product feature transformations are effective and interpretable, while generalization requires more feature engineering effort. With less feature engineering, deep neural networks can generalize better to unseen feature combinations through low-dimensional dense embeddings learned for the sparse features. However, deep neural networks with embeddings can over-generalize and recommend less relevant items when the user-item interactions are sparse and high-rank. In this paper, we present Wide & Deep learning-jointly trained wide linear models and deep neural networks-to combine the benefits of memorization and generalization for recommender systems. We productionized and evaluated the system on Google Play, a commercial mobile app store with over one billion active users and over one million apps. Online experiment results show that Wide & Deep significantly increased app acquisitions compared with wide-only and deep-only models. We have also open-sourced our implementation in TensorFlow.

No abstract

The reactions of hexachlorocyclotriphosphazatriene, N(3)P(3)Cl(6), with mono- (1 and 2) and bisferrocenyldiamines (3-5), FcCH(2)NH(CH(2))(n)NHR(1) (R(1) = H or FcCH(2)-), produce mono- (6 and 7) and spirocyclic bisferrocenylphosphazenes (8-10). The fully substituted phosphazenes (11-15 and 18-21) are obtained from the reactions of corresponding partly substituted phosphazenes (6-10) with excess pyrrolidine and NH(2)(CH(2))(3)ONa, respectively. The reactions of 6 with 1-aza-12-crown-4 afford geminal (16) and tris (17) crown ether-substituted phosphazenes. The structural investigations of the compounds have been verified by elemental analyses, mass spectrometry, Fourier transform IR, (1)H, (13)C, and (31)P NMR, and DEPT, COSY, HETCOR, and HMBC techniques. The crystal structures of 7, 10, 11, and 15 have been determined by X-ray crystallography. In 16 and 17, there are one and two stereogenic P atoms, respectively, and they are expected to be in enantiomeric mixtures. The structures of 18-21 look similar to a propeller. In 20 and 21, there are two stereogenic P atoms, and they exist as cis (meso; 20a and 21a) and trans (racemic; 20b and 21b) geometric isomers, according to the chiral solvating agent (S)-(+)-2,2,2-trifluoro-1-(9'-anthryl)ethanol experiments. Moreover, the compounds 18 and 19 have three stereogenic P atoms, and they exist as enantiomeric mixtures. Cyclic voltammetric investigations of compounds 6-21a reveal that ferrocene redox centers undergo oxidation concurrently at the same potential with basically reversible peaks, and these compounds appear to be quite robust electrochemically. The compounds 11-15 have been screened for antibacterial activity against gram positive and gram negative bacteria and for antifungal activity against yeast strains.The compounds 11, 12, 14, and 15 are evaluated for antituberculosis activity against reference strain Mycobacterium tuberculosis H37Rv (ATCC 27294). Interactions between compounds 11-15 and pBR322 plasmid DNA are studied by agarose gel electrophoresis. These compounds induce conformational changes in the DNA helix.

No abstract

No abstract