The Creation of Custom Data Sets to Meet Customer Needs: The BSC Project

Written by Caludia Gómez | 05/26/23

Rapidly advancing technology and the growing need for accurate and efficient data analysis have led organizations to seek customized data sets tailored to their specific needs. 

In this article, we will explore the creation of custom data sets containing bilingual segments classified by domain and style, using the Pangeanic BSC project as a key example.

 

What is a data set and what are the different types?

A data set is a structured collection of information, which can be numeric, textual, visual, or a combination of these data types. Data sets are used in various fields and disciplines, such as data science, artificial intelligence, statistics, scientific research, and many others, to perform analyses, studies, and experiments. Data sets can be divided into several categories depending on their type and structure.

Depending on the type of data

There are several types of data sets, which can be classified according to various characteristics, such as format, structure, and purpose. Some of the most common types of data sets according to their type are:

  1. Time series data: These are data sets that record the evolution of a variable over time. These data sets typically have associated timestamps, allowing for the analysis of patterns and trends over time. Examples of time series data sets include weather data, stock price data, and traffic data.

  2. Image data: These are data sets containing images, whether photographs, medical images, satellite images, or other types of images. These data sets are typically used in computer vision for object recognition and image analysis.

  3. Text data: These are data sets containing text, such as documents, text messages, tweets, or news. These data sets are used in natural language processing applications, such as sentiment analysis, text classification, and other text processing-related tasks.

  4. Social network data: These are data sets containing information generated by users on social networks, such as Facebook, Twitter, or Instagram. These data sets are used for social network analysis, opinion mining, and online behavioral studies.

  5. Geospatial data: These are data sets containing geographic information, such as GPS coordinates, maps, or geospatial sensor data. These data sets are used in mapping, location analysis, and geolocation applications.

These are just a few examples of the types of data sets that exist. Data sets can be very diverse and vary depending on the domain and purpose of analysis.

 

You might be interested in:

Working with aggregate data: What needs to be taken into account?

 

According to the structure of the data

Data sets can also be classified according to their structure. Some of the most common data types, based on their structure, include:

  1. Structured data: These are data sets that have a defined and organized structure, where the data is in a tabular format with rows and columns. Structured data is easy to analyze and process, as it usually has a predefined schema. Examples of structured data include databases, financial records and sales data.

  2. Unstructured data: These are data sets that do not have a defined structure and do not conform to a tabular format. This data is usually more difficult to analyze and process, as it may be in different formats, such as free text, images, videos, or audio files. Examples of unstructured data are text documents, images, videos, and social media data.

  3. Semi-structured data: These data sets have a partially defined structure. The data may contain information in different formats and have some organization, but does not comply with a completely defined structure like structured data. Examples of semi-structured data are XML documents, JSON files, and data in CSV format with optional fields.

  4. Hierarchical data: These are data sets that have a hierarchical structure, where data is organized in levels or layers. Hierarchical data is used in applications such as hierarchical databases, folder structures in file systems, and JSON-formatted data with object nesting.

  5. Graph data: These data sets are represented as graphs, where data are modeled as nodes and the relationships between them. Graph data is used for social network applications, network analysis, transportation routes, and complex relationships between entities.

 

The advantages of using a data set

Using a data set, which is a collection of organized and structured information, offers numerous advantages in a variety of contexts. Here are some important advantages of using a data set:

  1. Data-driven analysis and decision-making: A well-prepared and representative data set can provide valuable information for analysis and informed decision-making in a wide range of fields. Data can reveal patterns, trends, and correlations that can help to better understand a situation or problem, leading to better, evidence-supported decisions.

  2. Efficient research and knowledge acquisition: Data sets are fundamental tools for scientific research, academia, and knowledge gathering in general. They enable researchers and academics to efficiently collect, analyze and synthesize data to extract meaningful information, develop theories and validate hypotheses.

  3. Developing and training machine learning models: Data sets are essential for the development and training of machine learning models. These models use data to learn patterns and make predictions or classifications in a wide range of applications, such as image recognition, natural language processing, product recommendation, and more.

  4. Monitoring and tracking performance: The data sets are also useful for performance monitoring and tracking in a variety of areas, such as business performance, patient health status monitoring, weather and environmental tracking, and more. The data can be used to measure key performance indicators (KPIs) and evaluate progress toward established objectives.

  5. Identifying patterns and opportunities: Data sets can help identify patterns and opportunities that might otherwise go unnoticed. By analyzing large amounts of data, emerging trends, relationships, and opportunities can be uncovered, which can lead to the identification of new strategies, process improvements, and resource optimization.

  6. Personalizing and improving user experience: Data sets can also be used to personalize the user experience in digital applications and platforms. By collecting and analyzing data about users' preferences, behaviors, and needs, services, products or content can be personalized to provide a more relevant and engaging experience.

In short, data sets are fundamental tools for data analysis, research, machine learning development, and informed decision-making. They provide a solid foundation for decision-making, gaining insights, identifying patterns and opportunities, and improving the user experience, which can lead to better outcomes and greater understanding in a wide range of applications and contexts.

 

Read more:

The relationship between data science and machine learning

 

Uses of custom data sets

Personalized data sets enable companies to better understand their customers, allowing them to personalize product offerings and improve their customer experience.

Access to unique and customized data sets can provide organizations with a significant competitive advantage, enabling them to make informed decisions faster and more effectively.

Customized data sets can also provide valuable information on specific industries, helping organizations stay ahead of trends and developments. In addition, they can improve the performance of machine learning models by providing highly relevant and domain-specific data for training and validation.

Discover the Pangeanic BSC project

The Pangeanic BSC project focuses on the creation of customized data sets containing bilingual segments classified by domain and style. This innovative approach responds to the growing demand for high-quality customized data in various industries.

The project emphasizes bilingual data collection, which can be used to train machine translation systems, linguistic models, and other natural language processing applications. Data sets are classified by domain, ensuring that users can access data relevant to their industry and area of interest, leading to more accurate and meaningful results. In addition, stylistic classification allows for greater granularity of data, taking into account the specific nuances of different writing styles and registers.

In order to create a labeled bilingual English-Catalan data set, several steps were followed, as detailed below:

  1. Domain and text style selection: Fifteen different domains were carefully chosen covering a wide variety of topics, such as news, sports, technology, and health, among others. In addition, 7 different text styles were considered, such as formal news, informal blogs, social networks, forums, and others, to capture the diversity of text styles present on the web.

  2. Data source identification: Extensive web searches were conducted to identify relevant and reliable data sources for the selected domains and text styles. This included searching for websites, blogs, social networks, and forums that provide content in English and Catalan.

  3. Data crawling: A web crawling tool was used to obtain the data from the selected sources. Full web pages, documents, and social media posts were downloaded, and text was extracted in both English and Catalan in a systematic and automated manner.

  4. Data cleaning and processing: The data obtained was subjected to rigorous cleaning and processing to ensure quality and consistency. HTML tags were removed, formatting and spelling errors were corrected, and irrelevant or duplicate data was removed.

  5. Data validation and labeling: A thorough validation of the aligned data was performed to ensure its quality and accuracy. Possible alignment errors were reviewed and corrected. The data were then labeled with relevant metadata, such as font, domain, text style, and language, among others, to facilitate its use in future applications.

  6. Preparation of the data set: Finally, the data set was prepared and stored in a relational database with the respective metadata collected throughout the segment processing, for use in natural language processing applications.

 

Related content:

Tips for Creating Accurate and Useful Image Data Sets

 

 

Since representativeness in the construction of a text data set is essential to ensure the quality and reliability of the models that use them, some guidelines were followed to try to ensure this, classifying the text by domain and style. 

Special care was taken when selecting data sources so that they were varied, and to avoid bias in the data, as well as obtaining an adequate amount of data from different sources and writing styles to avoid over-representation of any of them. 

The representativeness of a data set is not static but can evolve over time. It is important to perform periodic updates of the data set, add new data from different sources and writing styles, correct possible errors in the annotation, and improve the quality of the data set.

In summary, an exhaustive process was undertaken that included selecting domains and text styles, identifying and obtaining data sources, data crawling, data cleaning and processing, data validation and labeling, and preparing the data set for use in natural language processing applications. This bilingual English-Catalan data set is a very valuable resource, especially considering that Catalan is a low-resource language.

By offering customized data sets that are tailored to client's unique needs, the Pangeanic BSC project sets a new standard for data quality and relevance, paving the way for more efficient and accurate data-driven solutions in a variety of industries.