KPI raw data(5,855,201 rows)

CDR raw data(9,232,275 rows)

Twitter API(44,989 tweets)

Keras: Open-source neural network library.The deep learning engineto train the model.

Bert: Pre-training language model.The embedding layer.

VaderSentimentSentiment analysisalgorithm for socialmedia content.The sentiment analysistool.

PostgreSQL:Open-source relationaldatabase managementsystem.Storing data in psql table.

Grafana: Open sourcemetric analytics &visualization suite.Visualize the anomaly results andtwitter sentiment analysis result.

Docker: A tool forbuilding and runningdistributed applications.Application deployment anenvironment version control.

Bert Embedding + LSTM87.1%

Keras Embedding + LSTM51.33%

Twitter APILive streaming data

VaderSentiment

Bert Embedding

Customer Issue Content(CSV file, 3 labels, 42,267 rows)

KPI STD data

LSTM

Reconstruction errorcalculation

Dimension Reduction AutoEncoder200 39 36 24 36 39

Data Anomalies

Output Shape39 Dimensions

Output Shape200 Dimensions

Input Shape39 Dimensions

Anomalies

As telecommunication companies seek to improve network coverage andmaintain customer satisfaction, they need to constantly monitorcustomer complaints and the status of their networks.This project, in collaboration with Tupl Inc. seeks to reduce the latencybetween the time a network coverage area experiences issues and whenthe network operator notices these issues. Deep learning models processseveral types of telecom data to predict which data are anomalies. - Call detail records (CDR) data from customer complaints determines if customers have recurrent cellular issues. - Key performance indicators (KPI) data determines which base stations are down at a given time. - Twitter tweets identify customers’ dissatisfaction on social media that isn’t directly reported to the company.

Sentiment Trends

Tweets locationwith sentiment value

SentimentDistribution

Issue labelDistribution

Detail of tweets

Anomaly station on topology map

Twitter Sentiment Analysis

Anomaly & Standard Deviation Heatmap

Anomalies

LSTM Layer(256 units)

DenseLayer(100 units, relu)

DenseLayer(3 units, softmax)

dropout = 0.1dropout = 0.1

CDR raw data(41 features)

2/3G data

4G VoLTE data

KPI-Like Data(5 features)

KPI-Like Data(8 features)

x1

x2

x3

x4

x5

x6

x7

x1

x2

x3

x4

x5

x6

x7

a1a1

a2

a3

a4

a5

a2

a3

a4

a5

a1

a2

a3

Reconstructionerror

2/3G Anomaly Result

4G Anomaly Result

Input Layer1432

Hidden Layer256

Output Layer1432716 716

Find Anomalies

Otsu ThresholdEncoder Decoder

Training Data

Input Shape(30, 768)

ClassificationResult

Telcom related tweet

Tweet with sentiment value

Comparing the Upload Traffic Volume, we plotted the input data in blue vs.the prediction data from our model in orange. We calculated the reconstruction error between the two datasets anddetermined the anomaly threshold for this feature to identify the anomalies.

Index of Site ID

IMSF

aultC

odeR

ate

PredictionIMSFaultCodeRate