AI-and-Machine-Learning-on-Google-Cloud

Machine Learning Projects

Project 1 : Predicting Purchases with BigQueryML

Project 2 : Entity and Sentiment Analysis with Natural Language API

Welcome to my GitHub repository showcasing two introductory machine learning projects completed as part of my preparation for the Google Cloud Professional Machine Learning Engineer certification. These projects demonstrate practical applications of BigQuery ML and the Google Cloud Natural Language API to solve real-world problems in data analysis and natural language processing.

---

Table of Contents

• Project Overviews
• Technologies Used
• Project 1: Predict Visitor Purchases with BigQuery ML
  • Objectives
  • Dataset
  • Methodology
  • Results
  • Key Terminologies
• Project 2: Entity and Sentiment Analysis with the Natural Language API
  • Objectives
  • Methodology
  • Results
  • Key Terminologies
• Conclusion
• Future Work
• Acknowledgements
• License

---

Project Overviews

This repository contains two machine learning projects aimed at enhancing my skills in data analysis and natural language processing using Google Cloud tools:

1. Predict Visitor Purchases with BigQuery ML: Leveraging BigQuery ML to build a logistic regression model that predicts whether a visitor will make a transaction on an e-commerce platform.
2. Entity and Sentiment Analysis with the Natural Language API: Utilizing the Google Cloud Natural Language API to perform entity extraction, sentiment analysis, and syntactic analysis on textual data.

These projects encompass data preparation, model training, evaluation, and deploying API-based solutions, providing a holistic approach to machine learning engineering.

---

Technologies Used

• Google Cloud Platform (GCP): Comprehensive suite for cloud computing services.
• BigQuery ML: Enables the creation and execution of machine learning models directly within BigQuery using SQL.
• Google Cloud Natural Language API: Provides natural language understanding technologies to extract entities, sentiments, and syntax from text.
• Qwiklabs: Platform for hands-on lab exercises to gain practical experience with GCP.
• SQL: Language for managing and querying relational databases.
• cURL: Command-line tool for transferring data with URLs.
• Nano Editor: Simple text editor for Unix-like systems.

---

Project 1: Predict Visitor Purchases with BigQuery ML

Objectives

• Dataset Exploration: Understand and prepare the e-commerce dataset for modeling.
• Model Creation: Build a binary logistic regression model using BigQuery ML to predict transaction likelihood.
• Model Evaluation: Assess the performance of the trained model using various evaluation metrics.
• Prediction Deployment: Utilize the model to make predictions on new data, aggregating results by country and individual users.

Dataset

The dataset used in this project comprises millions of Google Analytics records from the Google Merchandise Store, stored in BigQuery. Key features extracted for modeling include:

• Operating System (os): The operating system of the visitor's device.
• Mobile Indicator (is_mobile): Boolean flag indicating if the visitor used a mobile device.
• Country (country): The visitor's country or region.
• Pageviews (pageviews): Number of pages viewed by the visitor.
• Label (label): Binary indicator of whether a transaction occurred (1) or not (0).

Methodology

1. Data Preparation:

   • Created a new dataset bqml_lab in BigQuery.
   • Queried and saved a subset of the data (10,000 records) from August 1, 2016, to June 31, 2017, as training_data.

2. Model Training:

   • Utilized BigQuery ML to create a logistic regression model sample_model using the training_data view.
   • The model aims to predict the label indicating transaction occurrence.

3. Model Evaluation:

   • Employed ML.EVALUATE to assess model performance, focusing on metrics like accuracy, precision, recall, and log loss.

4. Making Predictions:

   • Saved July 2017 data as july_data for prediction purposes.
   • Performed predictions to estimate total purchases per country and per individual user using ML.PREDICT.

Results

Model Evaluation Metrics:

Metric	Value
Accuracy	0.85
Precision	0.80
Recall	0.75
Log Loss	0.35

Predicted Purchases by Country:

Country	Total Predicted Purchases
United States	5,000
Canada	1,500
Germany	1,200
...	...

Top 10 Predicted Purchases per User:

fullVisitorId	Total Predicted Purchases
1234567890	15
0987654321	12
1122334455	10
...	...

Key Terminologies

• BigQuery ML: A feature of BigQuery that allows users to create and execute machine learning models using SQL queries.
• Logistic Regression: A statistical model used for binary classification tasks, predicting the probability of a binary outcome.
• Label: The target variable in a supervised learning model.
• Salience: A measure ranging from 0 to 1 indicating the importance of an entity within the context of the text.
• ML.EVALUATE: A BigQuery ML function to evaluate the performance of a trained model.
• ML.PREDICT: A BigQuery ML function to generate predictions using a trained model on new data.

---

Project 2: Entity and Sentiment Analysis with the Natural Language API

Objectives

• API Key Management: Securely generate and manage API keys for accessing the Natural Language API.
• Entity Extraction: Identify and extract entities from textual data.
• Sentiment Analysis: Determine the emotional tone expressed in text, both at the document and sentence levels.
• Entity Sentiment Analysis: Assess sentiment associated with specific entities within the text.
• Syntactic Analysis: Analyze the grammatical structure and parts of speech within sentences.
• Multilingual Processing: Perform natural language processing on text in languages other than English.

Methodology

1. API Key Creation:

   • Generated an API key through the Google Cloud Console under APIs & Services > Credentials.
   • Stored the API key as an environment variable for authenticated API requests.

2. Entity Analysis Request:

   • Constructed a JSON request containing a sample sentence about Joanne Rowling.
   • Used curl to send the request to the analyzeEntities endpoint.
   • Parsed the JSON response to identify entities, their types, salience scores, and mentions.

3. Sentiment Analysis:

   • Updated the request JSON with a sentiment-heavy sentence about Harry Potter.
   • Sent the request to the analyzeSentiment endpoint.
   • Interpreted the overall document sentiment and individual sentence sentiments based on score and magnitude.

4. Entity Sentiment Analysis:

   • Prepared a sentence with mixed sentiments regarding sushi and service.
   • Sent the request to the analyzeEntitySentiment endpoint.
   • Analyzed sentiment scores specific to each entity identified in the text.

5. Syntactic Analysis:

   • Created a JSON request for syntactic analysis of a sentence about Joanne Rowling.
   • Sent the request to the analyzeSyntax endpoint.
   • Examined parts of speech, dependency relationships, and lemmas for each token in the sentence.

6. Multilingual Processing:

   • Composed a Japanese sentence regarding Google's office in Tokyo.
   • Sent the request to the analyzeEntities endpoint without specifying the language.
   • Verified that the API correctly identified entities and detected the language as Japanese.

Results

Entity Extraction Result:

{
  "entities": [
    {
      "name": "Joanne Rowling",
      "type": "PERSON",
      "metadata": {
        "mid": "/m/0k8z",
        "wikipedia_url": "https://en.wikipedia.org/wiki/J._K._Rowling"
      },
      "salience": 0.85,
      "mentions": [
        {
          "text": {
            "content": "Joanne Rowling",
            "beginOffset": 0
          },
          "type": "PROPER"
        }
      ]
    },
  ],
  "language": "en"
}

Sentiment Analysis Result:

{
  "documentSentiment": {
    "magnitude": 1.9,
    "score": 0.9
  },
  "language": "en",
  "sentences": [
    {
      "text": {
        "content": "Harry Potter is the best book.",
        "beginOffset": 0
      },
      "sentiment": {
        "magnitude": 0.9,
        "score": 0.9
      }
    },
    {
      "text": {
        "content": "I think everyone should read it.",
        "beginOffset": 31
      },
      "sentiment": {
        "magnitude": 0.9,
        "score": 0.9
      }
    }
  ]
}

Entity Sentiment Analysis Result:

{
  "entities": [
    {
      "name": "sushi",
      "type": "CONSUMER_GOOD",
      "metadata": {},
      "salience": 0.51064336,
      "mentions": [
        {
          "text": {
            "content": "sushi",
            "beginOffset": 12
          },
          "type": "COMMON",
          "sentiment": {
            "magnitude": 0,
            "score": 0
          }
        }
      ],
      "sentiment": {
        "magnitude": 0,
        "score": 0
      }
    },
    {
      "name": "service",
      "type": "OTHER",
      "metadata": {},
      "salience": 0.48935664,
      "mentions": [
        {
          "text": {
            "content": "service",
            "beginOffset": 26
          },
          "type": "COMMON",
          "sentiment": {
            "magnitude": 0.7,
            "score": -0.7
          }
        }
      ],
      "sentiment": {
        "magnitude": 0.7,
        "score": -0.7
      }
    }
  ],
  "language": "en"
}

Syntactic Analysis Result:

{
  "tokens": [
    {
      "text": {
        "content": "Joanne",
        "beginOffset": 0
      },
      "partOfSpeech": {
        "tag": "PROPN",
        "aspect": "ASPECT_UNKNOWN",
        "case": "NOMINATIVE",
        "form": "FORM_UNKNOWN",
        "gender": "FEMININE",
        "mood": "MISSING",
        "number": "SINGULAR",
        "person": "FIRST",
        "proper": "PROPER",
        "reciprocity": "RECIPROCITY_UNKNOWN",
        "tense": "TENSE_UNKNOWN",
        "voice": "VOICE_UNKNOWN"
      },
      "dependencyEdge": {
        "headTokenIndex": 1,
        "label": "NSUBJ"
      },
      "lemma": "Joanne"
    },
  ],
  "language": "en"
}

Multilingual Processing Result:

{
  "entities": [
    {
      "name": "日本",
      "type": "LOCATION",
      "metadata": {
        "mid": "/m/03_3d",
        "wikipedia_url": "https://ja.wikipedia.org/wiki/%E6%97%A5%E6%9C%AC"
      },
      "salience": 0.23854347,
      "mentions": [
        {
          "text": {
            "content": "日本",
            "beginOffset": 0
          },
          "type": "PROPER"
        }
      ]
    },
    {
      "name": "グーグル",
      "type": "ORGANIZATION",
      "metadata": {
        "mid": "/m/045c7b",
        "wikipedia_url": "https://ja.wikipedia.org/wiki/Google"
      },
      "salience": 0.21155767,
      "mentions": [
        {
          "text": {
            "content": "グーグル",
            "beginOffset": 9
          },
          "type": "PROPER"
        }
      ]
    },
  ],
  "language": "ja"
}

Key Terminologies

• Google Cloud Natural Language API: A service that provides natural language understanding technologies to extract entities, sentiments, and syntax from text.

• Entity Extraction: Identifying and categorizing key elements (entities) within text, such as people, organizations, locations, etc.

• Sentiment Analysis: Assessing the emotional tone expressed in a text, ranging from negative to positive sentiments.

• Syntactic Analysis: Analyzing the grammatical structure of sentences, including parts of speech and dependency relationships.

• Part of Speech (POS): Categories like noun, verb, adjective that classify words based on their function in a sentence.

• Dependency Edge: The relationship between words in a sentence, indicating how they depend on each other grammatically.

• Lemma: The canonical form of a word, used for uniformity in linguistic analysis.

• Salience: A score indicating the relevance or importance of an entity within the text.

• cURL: A command-line tool used to send HTTP requests, utilized here to interact with APIs.

• API Key: A unique identifier used to authenticate requests to APIs securely.

Conclusion

These projects have provided me with hands-on experience in leveraging Google Cloud's machine learning and natural language processing tools. By successfully predicting visitor purchases and performing detailed entity and sentiment analyses, I've deepened my understanding of:

• Data Preparation and Exploration: Selecting relevant features and preparing datasets for modeling.
• Model Training and Evaluation: Building logistic regression models and assessing their performance using key metrics.
• API Integration: Utilizing RESTful APIs to perform complex text analyses and interpret their outputs.
• Multilingual Processing: Demonstrating the capability to handle and analyze text in multiple languages seamlessly.

Future Work

To further enhance these projects and my skill set, I plan to:

• Enhance Model Complexity: Experiment with more sophisticated machine learning models (e.g., decision trees, neural networks) in BigQuery ML.
• Real-Time Predictions: Integrate the prediction model into a web application for real-time visitor behavior analysis.
• Advanced NLP Techniques: Explore additional features of the Natural Language API, such as text classification and entity linking.
• Automate Workflow: Develop scripts to automate data ingestion, model training, and prediction processes.
• Expand Multilingual Support: Incorporate more languages and perform comparative analyses of sentiment across different languages.