Implementation of deep convolutional neural network for bug detection and prioritization in app review analysis

App Review Analysis (ARA) is an emerging field in the paradigm of Big Data analysis largely due to the broad user base and the possible advantages of automatic extraction of information. In order to gain customer loyalty and to succeed in the app industry, it is important to automate the methods for

Project Title

Implementation of deep convolutional neural network for bug detection and prioritization in app review analysis

Project Area of Specialization

Artificial Intelligence

Project Summary

App Review Analysis (ARA) is an emerging field in the paradigm of Big Data analysis largely due to the broad user base and the possible advantages of automatic extraction of information. In order to gain customer loyalty and to succeed in the app industry, it is important to automate the methods for collecting requirements from wide corpus of data from app reviews and feedback. In order to gain customer loyalty and to succeed in the app industry, it is important to expedite the process of extracting specifications from wide number of datasets through App analysis. In this report, we articulate this issue as a multi-label approach and suggest a classifier model from App reviews for the automatic extraction of content issues. It is time-consuming and inefficient to dissect each person feedback, considering the value of consumer feedback for mobile app management and progression. It is critical for engineers to upgrade and sustain their applications in the dynamic mobile app industry to identify the important intentions of customers (e.g., new features needed) timely and reliably. In the dynamic mobile app market, it is especially important for developers to upgrade and sustain their applications. Sentiment analysis and expectation extraction from crowd feedback provide an incentive to constructively gather the intentions of app users, such as bug diagnosis and refinement of functionality. Sentiment is a Natural Language Processing (NLP) area that generates templates that aim to define speech properties and describe them. Text mining has been a crucial instrument to make sense of the data in an environment where we produce 2.5 quadrillion bytes of data every day. This has helped enterprises to obtain valuable insights and automated processes. The method of using deep learning and language processing to interpret feelings into specific attributes is Sentiment Analysis (SA), or subjectivity analysis. In the field of machine learning, it is regarded as one of the most common fields of study, since it offers a way to analyze and evaluate views expressed by any number of individuals. In this proposal we articulate this challenge as an issue of multi-label classification and suggest a classification model from application reviews for automatic detection and extraction of content issues. On top of Word2Vec embedding, we have proposed a Multi Label Convolutional Neural model. It may also be used as a grouping model for other customer input research platforms, provided the positive outcomes and relevance of this model. This study will be novel in terms of Prioritization where Public opinions are measured in the post-query framework to manage the degree of grievance and therefore prioritize the quality opinions  by implementing hyperparameters in quality CNN model . The new structure would benefit multiple organisations with less human intervention to ensure consistent service provision and user satisfaction.

Project Objectives

In order to gain customer feedback and to succeed in the app industry, the main necessity and goal of this proposal is to simplify the extraction method of requirements from broad corpuses of App review results. This research aims to automatically remove consumer performance issues from app feedback. Study of the App Analysis (ARA) aims to solve the dilemma of background coverage.

The basic concept is to develop a multi-label platform for

Classification evaluations to address the complexity of criteria and to identify them efficiently with thoughtful scrutiny of context. By analyzing the influence of the hyper-parameters on the results obtained, we will recommend an effective sentiment classification algorithm using convolutionary algorithms. We will verify whether the network's design and structure play a role in determining analysis performance. We will analyze the function and effects of the embedding’s used on the network's precision rate. Our Objective is to formulate a new approach to analyze big data obtained from blogs, social networks that train a convolution of neural networks inspired by the successful results produced by sentiment sentence validation data sets and report and if there's any change when using Word2Vec approach. This study will also prioritize the bugs or failures through Data retrieval from app reviews using classification and clustering that can help in identification of functional and nonfunctional requirements and their severity with time. 

We will systematically survey the automatic bug prioritization processes because of their success and significance. This article, in particular, offers a brief theoretical analysis for problem reports to motivate the need for bug prioritization work. The latest bug prioritization work and any future challenges in dealing with bug prioritization are outlined.

Project Implementation Method

The implementation framework relies on different phases, including the selection of app types, the collection of apps and the compilation of final feedback (Figure 1). We will classify the list of most successful Android apps in the first section of category selection. In the second step, applications with more than one million user ratings will be deemed eligible for preference from the collection of apps of the selected product categories. We will retrieve hundreds of the most important user feedback for each framework using URL as data. The collected repository of feedback is automatically classified to assess the consistency of the classification algorithm.

According to the characteristics of the standard consistency model, a label from the label package will be applied to each analysis using machine learning techniques. Reviews will be access to dissemination over class-labels for identification and consistency attributes (Figure 2). Now, together with a functional criteria group, we will explore the output of the convolutionary neural network that automatically classifies the reviews into the given categories of NFRs. Functionalities of simple CNN consists of the following convolution layer: called as embedding, convolution, pooling and completely connected via a differentiable function, that will translate the input volume into an output. The architecture consists mainly of two parts, such as details,

CNN procedure compilation, description and data method subsequently. In order to perform operations on sentence terms, user input will be translated into numeric form, transformed into incorporating vectors of floating data points that will initialize the CNN classification model. Further use pre-trained embedding from the specified dataset to classify user ratings into different categories will be used. In the model, pre-based embedding, trained on Google news datasets of billion terms, called word2vec, would be used to convert the input data into vectors, enhancing efficiency. The ultimate project plan consists of the following steps:

For multi-label ratings, a CNN classifier is deployed. Three CNN configurations architecture will initially be applied over a simulated dataset of user ratings, and findings for 10 epochs will be registered.  To test the CNN model's performance, a simulated dataset of 10,000 reviews will be used and findings will be recorded for three epochs. Using the accuracy parameters, the proposed model will be tested. The evaluation results would provide CNN with potentially improved results comparison to other methods that will provide a performance preference to our proposed model of implementation.

Benefits of the Project

The most critical feature of contemporary applications analysis is to be in touch with its customer base. Knowing just what buyers think about new and existing products or services, current campaigns, and customer care offers is essential for these businesses. Sentiment Analysis may help turn unstructured data easily into organized public perception data on goods, services, brands or any other subject on which people may give views. This knowledge can be very helpful for various applications such as brand research, business development, product reviews, etc. CNN models have subsequently been shown to be powerful for NLP and have obtained outstanding results in expression modeling and other typical NLP activities for textual parsing search query recovery. This simple model produces outstanding results on multiple datasets despite minimal tuning of classifiers. Our model will Provide application developers/team an easy platform to analyze their bugs. The use of AI techniques will help identify the major bugs in applications in fastest way possible The whole process is automated and hence more efficient than manual bug detection and prioritization process. This system will reduce the overall cost of company as no extra resource will be needed for this bug detection and prioritization process. Traditionally company bear the burden of extra resources who do these types of tasks for company Provide a medium to judge the overall quality of application. Even if the application is less bug prune, our system has ability to show insight for betterment of application Our system can detect both functional and non-functional issues in reviews. Our system has ability to predict future bugs based on current trend.

Technical Details of Final Deliverable

the final deliverable will be a software system based on   Proposed  algorithm which is influenced by functionality and is structured to replicate neurons' interaction patterns within the neural network. Inside a CNN, the neurons are separated into a three-dimensional structure, for each set of neuron evaluating area or function. In other terms, each neuron segment involves in one aspect of the picture being described. CNNs use the layer forecasts to create a final result that offers a vector of probability ratings to represent the whether each function belongs to a certain class.

A CNN is composed of several kinds of layers:

• Convolutionary layer will generate a function map by adding a filter that scans the entire text, to predict the class probabilities for each element.
• Pooling layer will decrease the amount of data produced for each function by the convolutionary layer and preserves the most crucial data (the process of the convolutional and pooling layers usually repeats several times).
• Completely linked input layer-"flattens" the results produced by previous layers to transform them into a feature object which can be used by the very next layer as a source.
• Completely linked layer will predict an effective mark applies labels over the feedback created by the function analysis. This layer will generate the final prospects to determine the output details

In this study the CNN infrastructure is a crucial factor in deciding its quality and success. The way the layers are organized, which materials are included in each layer, and how they are constructed will also determine the speed and precision at which different tasks can be done.

Final Deliverable of the Project

Software System

Core Industry

IT

Other Industries

Others

Core Technology

Artificial Intelligence(AI)

Other Technologies

NeuroTech, Big Data

Sustainable Development Goals

Decent Work and Economic Growth, Industry, Innovation and Infrastructure, Partnerships to achieve the Goal

Required Resources

If you need this project, please contact me on contact@adikhanofficial.com