DeepMood: Modeling Mobile Phone Typing Dynamics for Mood Detection

“y, DeepMood provides a late fusion framework. It first models each view of the time series data separately using Gated Recurrent Unit (GRU) [12], a simplified version of Long Short-Term Memory (LSTM) [24]. It then fuses the output of the GRU from each view. As the GRU extracts a latent feature representation out of ea…”

“• XGB: The implementation of a tree boosting system from XGBoost6 [10] is used.”

“A typical GRU is formulated as: rk = sigmoid(Wr xk + Ur hk−1) zk = sigmoid(Wzxk + Uzhk−1) h̃k = tanh(Wxk + U(rk ⊙ hk−1)) hk = zk ⊙ hk−1 + (1 − zk ) ⊙ h̃k (3) where ⊙ is the element-wise multiplication operator, a reset gate rk allows the GRU to forget the previously computed state hk−1, and an update gate zk balances…”

“There are 7 participants with bipolar I disorder that involves periods of severe mood episodes from mania to depression, 5 participants with bipolar II disorder which is a milder form of mood elevation, involving milder episodes of hypomania that alternate with periods of severe depression, and 8 participants with no…”

“Inspired by the great success of deep RNNs in the applications of other sequence tasks, including speech recognition [20] and natural language processing [4, 34], in this work, we propose a deep architecture to model complex event sequences of mobile phone typing dynamics.”

“The GRU has been attracting great attentions since it overcomes the vanishing gradient problem in traditional RNNs and is more efficient than the LSTM in some tasks [14].”

“Inspired by the great success of deep RNNs in the applications of other sequence tasks, including speech recognition [20] and natural language processing [4, 34], in this work, we propose a deep architecture to model complex event sequences of mobile phone typing dynamics.”

“Based on the Kolmogorov-Smirnov test, for 98.06% in HDRS and 99.52% in YMRS of the distribution pairs, we can reject the null hypothesis that two samples are drawn from the same distribution with significance level α = 0.01.”

“Since RNNs in such a form would fail to learn long term dependencies due to the exploding and the vanishing gradient problem [5, 23], they are not suitable to learn dependencies from a long input sequence in practice.”

“Note that a dropout layer [22] is applied before feeding the output from GRU to the fusion layer which is a regularization method designed to prevent co-adaptation of feature detectors in deep neural networks.”

“In the first stage, each 2This is for privacy concerns, because malicious person may be able to unscramble and recover the texts using such information. view of the time series is separately modeled by a Recurrent Neural Network (RNN) [34, 44].”

“A typical GRU is formulated as: rk = sigmoid(Wr xk + Ur hk−1) zk = sigmoid(Wzxk + Uzhk−1) h̃k = tanh(Wxk + U(rk ⊙ hk−1)) hk = zk ⊙ hk−1 + (1 − zk ) ⊙ h̃k (3) where ⊙ is the element-wise multiplication operator, a reset gate rk allows the GRU to forget the previously computed state hk−1, and an update gate zk balances…”

“Since RNNs in such a form would fail to learn long term dependencies due to the exploding and the vanishing gradient problem [5, 23], they are not suitable to learn dependencies from a long input sequence in practice.”

“Based on the Kolmogorov-Smirnov test, for 98.06% in HDRS and 99.52% in YMRS of the distribution pairs, we can reject the null hypothesis that two samples are drawn from the same distribution with significance level α = 0.01.”

“Distance based methods include K nearest neighbor classifier [16, 26, 27, 38, 47, 49] and SVMwith local alignment kernel [32, 41, 42] by measuring the similarity between a pair of sequences.”

“Following the idea of Multi-view Machines [9], Factorization Machines [39], or in a conventional fully connected fashion, three alternative fusion layers are designed to integrate the complementary information in the multi-view time series to produce a prediction on the mood score.”

“Feature based methods first transform a sequence into a feature vector and then apply conventional classification models [1, 25, 30, 31, 52].”

“Distance based methods include K nearest neighbor classifier [16, 26, 27, 38, 47, 49] and SVMwith local alignment kernel [32, 41, 42] by measuring the similarity between a pair of sequences.”

“Distance based methods include K nearest neighbor classifier [16, 26, 27, 38, 47, 49] and SVMwith local alignment kernel [32, 41, 42] by measuring the similarity between a pair of sequences.”

“Distance based methods include K nearest neighbor classifier [16, 26, 27, 38, 47, 49] and SVMwith local alignment kernel [32, 41, 42] by measuring the similarity between a pair of sequences.”

“provide a brief survey on the sequence prediction problem where sequence data are categorized into five subtypes: simple symbolic sequences, complex symbolic sequences, simple time series, multivariate time series, and complex event sequences [50].”

“Distance based methods include K nearest neighbor classifier [16, 26, 27, 38, 47, 49] and SVMwith local alignment kernel [32, 41, 42] by measuring the similarity between a pair of sequences.”

“Distance based methods include K nearest neighbor classifier [16, 26, 27, 38, 47, 49] and SVMwith local alignment kernel [32, 41, 42] by measuring the similarity between a pair of sequences.”

“Continuous real-time monitoring in naturalistic settings and collection of automatically generated smartphone data that reflect illness activity could facilitate early intervention and have a potential use as objective outcome measures in efficacy trials [2, 6, 18].”

“To the best of our knowledge, no previous studies [3, 19, 21, 37, 40, 46] have investigated the relationship between mobile phone typing dynamics and mood states.”

“80 is spent on inpatient care, suggesting early intervention and improved prevention management could decrease the financial impact of this illness [35].”

“Feature based methods first transform a sequence into a feature vector and then apply conventional classification models [1, 25, 30, 31, 52].”

“Study findings may be unreliable when using rating scales as outcome measures due to methodological issues such as unblinding of raters and patients, differences in rater experiences and missing visits for outcome assessments [15, 18, 36].”

“To the best of our knowledge, no previous studies [3, 19, 21, 37, 40, 46] have investigated the relationship between mobile phone typing dynamics and mood states.”

“Study findings may be unreliable when using rating scales as outcome measures due to methodological issues such as unblinding of raters and patients, differences in rater experiences and missing visits for outcome assessments [15, 18, 36].”

“To the best of our knowledge, no previous studies [3, 19, 21, 37, 40, 46] have investigated the relationship between mobile phone typing dynamics and mood states.”

“Continuous real-time monitoring in naturalistic settings and collection of automatically generated smartphone data that reflect illness activity could facilitate early intervention and have a potential use as objective outcome measures in efficacy trials [2, 6, 18].”

“Feature based methods first transform a sequence into a feature vector and then apply conventional classification models [1, 25, 30, 31, 52].”

“Distance based methods include K nearest neighbor classifier [16, 26, 27, 38, 47, 49] and SVMwith local alignment kernel [32, 41, 42] by measuring the similarity between a pair of sequences.”

“To the best of our knowledge, no previous studies [3, 19, 21, 37, 40, 46] have investigated the relationship between mobile phone typing dynamics and mood states.”

“Model based methods assume that sequences in a class are generated by an underlying probability distribution, including Naive Bayes [11], Markov Model [51] and HiddenMarkov Model [43].”

“propose to fuse multi-view data through the operation of tensor product and assume that the effects of feature interactions across views have a low rank [7, 9].”

“Model based methods assume that sequences in a class are generated by an underlying probability distribution, including Naive Bayes [11], Markov Model [51] and HiddenMarkov Model [43].”

“Distance based methods include K nearest neighbor classifier [16, 26, 27, 38, 47, 49] and SVMwith local alignment kernel [32, 41, 42] by measuring the similarity between a pair of sequences.”

“Feature based methods first transform a sequence into a feature vector and then apply conventional classification models [1, 25, 30, 31, 52].”

“There are also some work incorporating multiple views into the process of subgraph mining [8] and deep learning [54] to help identify meaningful patterns from data.”

“There are also some work incorporating multiple views into the process of subgraph mining [8] and deep learning [54] to help identify meaningful patterns from data.”

“Here, success of newly developed models is primarily reported through performance measures of the ‘accuracy of the (best) classifier’ or ‘robustness of clustering’ [2, 4, 6, 23, 27, 33, 44, 45, 50, 57, 61, 62, 63, 64, 67, 86, 89, 112, 122, 128, 133, 134, 135, 136, 139, 141, 144, 145, 152, 153, 155, 165, 168, 176, 177,…”

“They have also been used to monitor PD based on motor movements [115, 116], or to detect depressive states based on typing dynamics [90].”