EquitLabor

Automated labor abuse detection and condition monitoring based on social media via deep neural networks for natural language processing.

In the context of a rapidly industrializing world, international sources of labor have become an increasingly accessible option for major corporations and businesses. Through this workforce, labor is provided at an exceedingly low cost, allowing for maximum yield and profit while maintaining minimal expenses. However, the price of these high yields for corporations is a confluence of non-compliance with labor regulations and gross abuse of lower-income workers. This issue holds to be prevalent in the developing world and Asia, which yields $51 billion US dollars in illegal profits per year. China is an area that hosts the prime conditions for this form of illicit labor; it has seen major economic growth and rapid industrialization in past years. The labor force contains approximately 290 million “peasant” workers, constituting 35% of the total working class, which accounts for the success seen by the country. Prioritization of economic growth over the working class’ well-being has resulted in poor working conditions, low wages, and virtually no union protections. 

The problem also has global implications, as many of the commodities manufactured in China are exported to other nations within the global networks of trade. The pressure to produce goods at low costs in order to compete within the global market has only exacerbated this issue. Despite the high prevalence of worker abuse and exploitation, identifying the instances of worker rights violations in supply chains has proven to be difficult due to the sheer size of international supply networks and the active suppression of worker solidarity. Limited transparency in global supply chains and the perverse sourcing of labor have contributed to the continuation of large-scale labor exploits by corporations and an overall lack of accountability. 

Exploited workers often lack access to legal means necessary to ensure their rights are being protected, to bargain for higher quality working conditions, and to attain fair wages. This can lead to socio-economic issues for laborers: including poverty, lack of access to education, inadequate healthcare, and limited opportunities for upward mobility. The perpetuation of worker exploitation leads to a cycle of poverty that is passed down through generations.

We are utilizing deep neural networks for natural language processing to detect and provide an analysis of labor abuse based on first-hand accounts from social media. 

Our program extracts posts from various social media platforms that may contain information pertaining to labor conditions or potential cases of exploitation. Our model then uses natural language processing to analyze the posts to determine instances of labor abuse. After we detect exploitation, we will send a detailed report regarding the incident to the respective HR departments of companies who are sourcing labor from the site we have observed; therefore alerting them of the situation. Our data can also be used in legal proceedings as evidence in court to support claims of labor abuse. 

To retrieve data from social media platforms, we at EquitLabor have developed a web scraping program that uses social media APIs to extract comments and posts directly from the corresponding platforms. We are using Python to build our model, to efficiently analyze big data, and to perform data preprocessing techniques. We have developed a deep neural network model that performs sentiment analysis, text classification, and named entity recognition (NER) to determine cases of labor exploitation.

EquitLabor aims to protect the rights of exploited low-income workers in the Eastern Hemisphere, such as the Asia-Pacific region where abuse is prevalent. The individuals who fall victim to abuse and exploitation from contracting companies are of vulnerable status. This demographic typically includes students, the poor, and even children in various cases. In their local areas, there is often little action made by regional authorities or governments to address these issues. These workers account for an immense portion of the exports from their respective sectors each year; their manufacturing capabilities have “boosted” the national economy. 

Within China, we are zeroing in on the state of worker rights in Shenzhen: a rapidly growing hub for manufacturing and tech production. Shenzhen is a Special Economic Zone, which creates vast incentives for investment from foreign parties. The region has seen a rise in GDP per capita of 24,569% between the years 1978-2014, greater than that of Hong Kong and Singapore. This growth has been fueled by the introduction of the “peasant” worker class; which consists of internal migrants from China’s countryside. In light of rapid industrialization, a large portion of the young rural population has sought economic opportunity and employment in the city due to scarce opportunities to make ends meet in the countryside. Despite physically residing and operating within the city, they are often not granted the same range of benefits as native residents, forming a dependency upon their employers which gives them monopolizing power over workers. This vulnerability is dangerous as it makes workers in this region more susceptible to exploitation. The jobs that are taken on by this population pay extremely low wages that are typically only enough to sustain life. In such positions, workers often exceed legal working hours as defined by Chinese Labor Laws, as they are coerced to work overtime most days of the week. Despite such violations of regulations, employers receive little to no repercussions as the local government doesn’t often take action in the interest of economic endeavors. 

In 1993, a fire at a Shenzhen factory killed 81 toy factory workers. There was high demand from workers to reform, but a similar incident occurred once again in 2013 at a poultry plant; this revealed that the same violations had taken place. Even amidst almost two decades of industrial and social progress, the minimal protections for laborers remained relatively the same. Although discontent runs high, workers find it difficult to initiate change. They are isolated from any means to do so, and even when an opportunity to advocate change comes forth, there is minimal fruition from their attempts due to authorial apathy and restrictions. Social media has proven to be an effective outlet for these workers to voice their complaints. In 2020, a group of workers at a factory in southern China that supplies Apple used the platform Weibo to discuss the poor working conditions and low wages they were given. This incident received widespread media coverage and led to considerable improvements in working conditions at the factory. In more isolated cases, workers' complaints will go unnoticed in the limitless array of information available on social media. Our solution addresses this problem by identifying these cases that previously would have been lost and through this, we are able to amplify these workers' messages. As awareness allows consumers and companies to be more conscious of where and how labor is being sourced, we are able to directly impact the circumstances in which these laborers work.

We have collaborated on a research paper regarding the exploitation of Bangladeshi women in the ready-made garment sector. The paper details the systemic worker abuse that has been upheld by corporations’ interest in profitable labor conditions. These conditions are characterized by poor economic and social conditions. We expressed prolonged interest in this topic as it is an issue that we, as consumers, directly impact. We performed extensive research on supply chain transparency and sustainability, providing us with a solid understanding of the needs that must be addressed through our program. Additionally, we have strong background knowledge in various coding languages. Since Python is the preferred language for performing big data analytics and building machine learning models, we decided to use it when developing our program. As a team, we are proficient in programming in C, C++, Python, and Java, as well as Windows and Linux OS Security and Management.   

Our team member, Arya, had previous work experience which inspired us to fight against this global issue. Over the summer, he taught programming to children interested in technology and was originally offered an unpaid internship. However, many of his coworkers in the same position as him were being paid. That’s when Arya realized that many companies look for possible cuts they can make to their spending, especially to employee wages. The consideration of the circumstances of workers in developing nations combined with our backgrounds in technology sparked our interest in a tech-powered solution that could help fight the widespread issue of labor abuse. We at EquitLabor have made it our mission to fight for the rights of workers in places where labor abuse is prevalent and left individuals in a perpetual state of poverty.

Due to the barriers found with contacting manufacturing workers in Shenzhen, we opted to reach out to legal agencies to better understand how legal aid can benefit laborers. We conducted interviews with local lawyers to gain their perspectives on issues regarding labor abuse and current barriers to its prosecution. We learned that large corporations have complex global supply chains, which can make it convoluted and hard to identify and track instances of labor abuse. This lack of transparency also makes it harder for prosecutors to build a case against the company. The lawyers that were interviewed revealed that they believe NGOs and other groups that are concerned with labor rights may not have enough resources to investigate and effectively document incidents, making it difficult to produce cases against large corporations. Our program is based on these considerations and has been developed to address these issues.

We hope that within the next year, we can pilot our program in Shenzhen to receive feedback from NGOs and consult laborers on improving our program. 

Our solution provides a new and significantly improved approach to tackling the problem of labor abuse and exploitation by utilizing cutting-edge technology to detect and analyze labor conditions. By leveraging deep neural networks for natural language processing, our program is able to evaluate and isolate potentially serious cases of labor abuse. 

This approach allows for more efficient and accurate detection of labor abuse, as it can rapidly analyze large amounts of data and identify indicators of worker exploitation. Cases of labor abuse may be missed by traditional methods such as social and labor audits conducted by companies, which are incentivized to hide malicious activities regarding labor abuse. Report-reliant systems may prove ineffective as they are often performed by the companies themselves or by third-party auditing firms that the companies hire. This creates a conflict of interest, as the auditing firm or company may be encouraged to present a positive image of the company's labor practices to maintain its reputation or avoid regulatory penalties. Additionally, these inspections are often only conducted periodically, which means that any labor violations that occur outside of the audit period may go undetected. In regard to the method of inspections, companies may be able to manipulate this process by hiding evidence of labor violations or by presenting a sanitized version of their operations to the inspectors. 

In contrast, EquitLabor sources first-hand reports of labor abuse and avoids the potential corruption of companies attempting to hide their immoral labor practices. We do this  by completely removing companies' involvement in the process of identifying exploitation. Our fully-automated program will also function daily in order to provide real-time incident reporting and condition monitoring. Manufacturing laborers are typically neglected due to their low socio-economic status and lack of access to resources that could aid their situation. Our solution gives laborers a voice by amplifying their complaints.

Our program can also provide detailed information about specific instances of labor abuse, which can be used as evidence in legal proceedings. This can help to increase the chances of successful legal action against abusive employers and systemic uses of exploitative labor. By identifying personal accounts of abuse that may go unnoticed in the vast ocean of social media, we give a voice to people’s reports that would have disappeared in the endless stream of information circling the internet. Overall, our novel approach to labor abuse detection by using first-hand accounts can result in a greater level of accountability for employers and help deter future instances of labor abuse.

EquitLabor will be catalytic in raising awareness, increasing supply chain transparency, improving monitoring, and providing evidence to hold corporations responsible. Our program can provide insight and evidence of labor abuse that can be used to call for action from the public, policymakers, and businesses by exposing unsustainable labor practices and facilitating investigations of abuse. EquitLabor’s ability to analyze text data can help to provide a deeper understanding of the context and dynamics of labor abuse, which can be used to inform decisions about policy and business practices.

Due to local restrictions imposed upon workers in advocating for their rights, and bargaining for better working conditions, minimal change results from such attempts. A prospective solution comes in with informing consumers on the purchasing end of the situation. By alerting consumers of the exploitation and abuse behind their products, it may result in more thoughtful, and conscience-driven purchases. For example, a fall in Apple sales resulted in a cut in the employee count and an increase in overtime hours. As a result, many employees of manufacturing employers quit, forcing employers to implement better working conditions in an attempt to maintain their workforce. As a general principle, through awareness of labor conditions for workers in areas such as Shenzhen, companies and consumers become more cautious in their approach. We aim to accomplish this awareness next year by accumulating large amounts of labor abuse instances and compiling a report that details the violations by companies in the sector. 

At EquitLabor, we are working on a web scraping program that utilizes APIs to gather comments and posts from various social media platforms. This program will allow us to extract the necessary data from these platforms efficiently. We have already created a functioning program to extract data from Reddit. This can be easily altered to accommodate other social media platforms such as Sina Weibo, which is predominantly used in Shenzhen as the major social media platform. Previously, the platform allowed workers from various manufacturing districts in China to share information and organize protests and strikes; this has helped bring attention to issues such as low wages, long working hours, and poor working conditions in the factories. To ensure the quality of our data we are using data preprocessing techniques such as data cleaning, dimensionality reduction, and feature engineering. For our prototype, we have acquired a smaller dataset featuring posts from Reddit to test our program and model. We are utilizing the Python libraries, Pandas and Numpy, to convert our raw data into a DataFrame in order to easily manipulate and analyze the information. 

To detect instances of labor exploitation, we have created a deep neural network model that performs sentiment checks, text classification, and named entity recognition techniques (NER). This model allows us to analyze text data and identify any relevant complaints that may indicate abuse. To perform anomaly detection, we are using long short-term memory networks (LSTMs), a type of recurrent neural network (RNN), and employing statistical methods such as the Empirical Rule (Three-Sigma Method). We will account for language differences in our data by incorporating machine translation processes. An evaluation of our model will be conducted using a confusion matrix to calculate evaluation metrics such as accuracy, precision, recall, and F1-score.

Currently, we haven’t launched our solution yet. We are still in the process of refining and optimizing our program. However, our goal is to serve workers facing abuse in regions with widespread internet access. The target population that we hope to directly impact when we pilot our solution is factory workers in Shenzhen. We are choosing Shenzhen as our target population because it is the top manufacturing center for tech products in China. According to a Shenzhen Internet Development Research Report, the city has an internet penetration rate of around 77%, demonstrating that the majority of the city and even the 2 million active workers in the manufacturing sector of the city  have access to the internet and social media sites—including Weibo. An estimated 1.5 million laborers could benefit from our program. Our goal is to directly impact 40,000 workers within the next year. The International Labour Organization (ILO) estimates that around 21 million workers in the Asia-Pacific region are exploited in some capacity. With increased funding, access to resources, and partnerships with NGOs, we hope to expand to greater regions in Asia and benefit those populations as well.

We experience barriers when pulling large amounts of data from various social media platforms. In some cases, such as Sina Weibo, direct downloading of large data sets is not possible. To combat this we are breaking data requests into small chunks and combining the results. However, in order to handle this large amount of data being regularly collected through social media, we will need a server powerful enough to handle our program, the computational demands of our machine learning model, store the generated data, handle requests to our website, and generate day-to-day reports in order to send emails alerting respective companies of labor exploitation detected in their workforces. Using a cloud service such as Amazon Web Services (AWS), which provides powerful virtual servers that can be configured to handle our demands and are easily scalable, may be a viable option. However, the price required to buy a high-performance AWS product can range anywhere from $80 to $400 per month, and we currently lack the finances to allocate the necessary funds to host our servers. 

Our program revolves around the utilization of social media in order to identify instances of worker abuse. Social media provides a platform for workers in order to voice their grievances and amplify their complaints. However, workers face restrictions on such platforms due to censorship from government agencies that deem many complaints to be “sensitive” content. For example, on International Women’s Day in 2018, a popular account that garnered approximately 180,000 followers was abruptly deleted with no follow-up explanation for such action by the government. Nonetheless, social media still serves as an ideal platform for workers to express their discontentment with the current conditions as even heavier restrictions are imposed upon local labor unions. Under the Charity Law and Foreign NGO Law, the actions and funding of domestic and foreign organizations that aim to regulate worker rights are strictly monitored and restricted. For example, the Foreign NGO Law prohibits actions that threaten “China’s national unity, security, or ethnic unity” or damage “national interests, or societal public interest." These regulations fail to clearly define violations, opting to loosely define them in order to allow for loose interpretations. Attempts to suppress labor solidarity are veiled under the guise of maintaining national unity. EquitLabor aims to work around such tight restrictions imposed by the local government as we utilize APIs as our method of data collection. The APIs we have acquired allot access to posts that have been deleted, allowing us to consider content that may have been previously lost to censorship from government regulations. 

We currently do not have any partnerships. We hope to eventually collaborate with non-government organizations to spread awareness regarding labor exploitation and with law agencies to hold corporations accountable for their labor practices. We at EquitLabor believe that we can have an even more significant impact if it's part of a larger initiative that includes other ways of investigating labor abuse and if it's used in combination with other tools such as audits and inspections. 

We provide value to the populations we serve by giving workers a platform to voice their complaints against malicious company practices, with the goal of helping workers receive reasonable wages and a safe working environment. We hope to provide this by gathering first-hand accounts on social media and evaluating large concentrations of negative posts about a company to detect potential cases of labor abuse and report it to the company’s HR. Our primary customer segments are NGOs, news outlets, and auditors. We will also disclose our data to organizations sharing similar goals that can help take legal action against responsible corporations. We can also generate media attention by broadcasting our findings via news outlets. Our data can also be utilized by auditing firms, such as FLOCERT, which is the main enterprise responsible for certifying fair trade. 

Funding for EquitLabor will come from a combination of sources including grants, service contracts, and investment capital. We will apply for grants from foundations, non-profit organizations, and government agencies that support initiatives related to labor rights and human rights. We can also offer our services to government agencies and international organizations that are interested in detecting and addressing labor abuse. Another potential source of funding would come from seeking investment capital from venture capitalists, angel investors, and impact investors who are interested in supporting socially responsible businesses. Forming partnerships with companies and organizations that are interested in improving their labor practices and reputation would also be beneficial. We could also license our program to auditing firms such as FLOCERT.

EquitLabor will use a social enterprise model to generate revenue while also achieving our social mission. Will will seek sponsorships from organizations or businesses that share our mission and goals, in order to cover the costs of providing the service. Using crowdfunding platforms to raise funds from a large number of individuals who are interested in supporting its mission could also be beneficial. EquitLabor does not require large amounts of funding, our primary expenses will go toward operating our cloud servers.