使用机器学习衡量包容性创新的供求

The Initiative on the Digital Economy at the Massachusetts Institute of Technology runs a global Inclusive Innovation Challenge (IIC). The challenge brings together companies from around the world that are innovating in four categories — Skills Development & Opportunity Matching, Income Growth & Job Creation, Technology Access, and Financial Inclusion — so that the world’s lowest means populations can participate in and prosper from the digital economy.

麻省理工学院的数字经济倡议发起了全球性的包容性创新挑战(IIC) 。 挑战汇集了来自世界各地的公司，这些公司在四个类别(技能开发与机会匹配，收入增长与创造就业机会，技术获取和金融包容性)方面进行创新，从而使世界上最低的人群能够参与到数字化行业并从中受益经济。

We, at the MaRS Discovery District, evaluated the global demand and supply of services that address each area of inclusive innovation. Our goal was to identify the priority needs in different countries, and where there are gaps in meeting those needs. This publication will walk you through our methodology.

我们在MaRS探索区评估了解决包容性创新各个领域的全球服务需求和供应情况。 我们的目标是确定不同国家的优先需求，以及在满足这些需求方面存在差距的地方。 该出版物将引导您了解我们的方法。

衡量对包容性创新的需求 (Measuring Demand for Inclusive Innovation)

Let’s begin with measuring the demand for inclusive innovation. We calculated country-level scores for each of the four award categories. This allowed us to calculate country-level scores in order to rank countries according to their performance in each category. As we proceed, we’ll focus on Financial Inclusion to illustrate our methodology.

让我们从衡量包容性创新的需求开始。 我们计算了四个奖项类别中每个国家/地区的分数。 这使我们能够计算国家/地区分数，以便根据国家/地区在每个类别中的表现对其进行排名。 在继续进行过程中，我们将重点介绍“金融普惠”，以说明我们的方法。

We develop these category scores by aggregating data on a variety of underlying factors. The frameworks we use are based on frameworks developed by various international organizations. For financial inclusion, we use the G20 Financial Inclusion Indicators framework, which considers factors related to three dimensions of financial inclusion — the access, usage, and quality of financial services in a given country.

我们通过汇总各种潜在因素的数据来开发这些类别分数。 我们使用的框架基于各种国际组织开发的框架。 对于金融普惠，我们使用G20金融普惠指标框架，该框架考虑了与金融普惠的三个维度相关的因素-给定国家/地区的金融服务的获取，使用和质量。

The financial inclusion sub-indicator data come from the World Bank. For example, we use the number of ATMs per 100,000 people as an access indicator within the financial inclusion index. In turn, we use the percentage of the population with a bank account as a usage indicator.

金融普惠子指标数据来自世界银行。 例如，我们使用每10万人的自动柜员机数量作为金融普惠指数中的访问指标。 反过来，我们将拥有银行帐户的人口百分比用作使用指标。

Below, we show the regions around the world that are most in need of inclusive innovation in financial services. Lightest regions score lowest according to the financial inclusion index and are therefore considered to have the highest demand for solutions. Most African countries, for instance, have low financial inclusion scores, as well as parts of central, south, and southeast Asia. Many people in these areas don’t use traditional financial services like deposit accounts, or digital payments, partly because they have limited access to banks and even basic services like the Internet.

下面，我们显示了世界上最需要金融服务的包容性创新的地区。 最轻的地区根据金融普惠指数得分最低，因此被认为对解决方案的需求最高。 例如，大多数非洲国家以及中亚，南亚和东南亚的部分地区的金融普惠指数都较低。 这些地区的许多人不使用传统的金融服务，例如存款帐户或数字支付，部分原因是他们对银行乃至互联网等基本服务的访问权限有限。

衡量包容性创新的供应(Measuring Supply for Inclusive Innovation)

Next, we measure the supply of inclusive innovation around the world. We do this by estimating the number of companies in developing solutions in each of the four areas of inclusive innovation around the world.

接下来，我们衡量全球包容性创新的供给。 为此，我们估算了全球包容性创新四个领域中每个领域开发解决方案的公司数量。

By considering the supply of inclusive innovators alongside the demand for inclusive innovation in each of the four challenge areas, we hope to identify where there is an unmet need for solutions. This is the starting point for building a case to incentivize inclusive innovation in specific regions.

通过在四个挑战领域中的每一个领域都考虑包容性创新者的供应以及包容性创新的需求，我们希望确定哪里存在未满足的解决方案需求。 这是在特定地区建立激励包容性创新案例的起点。

Let’s begin by understanding the supply of inclusive innovators who have applied to MIT’s Inclusive Innovation Challenge. We looked at 1230 IIC applicant companies from 2017 to 2019. Latin America had proportionally more innovators than the other regions and the Skills Development & Opportunity Matching was the most popular award category in most regions except Africa and Latin America, where it was surpassed by Income Growth & Job creation. Financial Inclusion represented 15% of the applicants included in our analysis — this is 189 applicants.

首先，让我们了解适用于麻省理工学院的“包容性创新挑战”的包容性创新者的供应。 我们研究了2017年至2019年的1230家IIC申请人公司。拉美地区的创新者比例高于其他地区，技能开发和机会匹配是除非洲和拉丁美洲(在收入和收入方面都超过了)以外的大多数地区最受欢迎的奖项类别成长与创造就业机会。 金融普惠性代表了我们分析中15％的申请人-这是189个申请人。

Below is the geographical distribution of these companies over the demand map from above. As applicants were encouraged to select a region where their work is having the greatest impact, this map is able to show where solutions are being applied. For example, a company that is based in Amsterdam but is working to address Financial Inclusion in Kenya would appear as a dot in Kenya on the map.

以下是这些公司在上方需求图中的地理分布。 由于鼓励申请人选择对其工作产生最大影响的区域，因此该地图可以显示在哪里应用解决方案。 例如，一家设在阿姆斯特丹但致力于解决肯尼亚的金融普惠性的公司将在地图上显示为肯尼亚的一个点。

How can we identify companies that did not apply to the challenge but are addressing problems in financial inclusion? Using company descriptions taken from either Crunchbase, LinkedIn, the company’s website, or the text fields from the IIC applications, we can begin to identify the keywords that define the category.

我们如何确定不适用挑战但正在解决金融普惠性问题的公司？ 使用从Crunchbase，LinkedIn，公司网站或IIC应用程序中的文本字段获取的公司描述，我们可以开始识别定义类别的关键字。

Comparing the descriptions of companies under financial inclusion in contrast with companies in the other award categories allows us to further refine these words so that we can better identify the words that are most salient to Financial Inclusion.

与其他奖项类别中的公司相比，将金融包容性公司的描述进行比较可以使我们进一步完善这些用语，以便我们可以更好地识别出最适合金融包容性的词语。

The word cloud above shows us the words financial inclusion companies used to describe themselves. The size of the word corresponds to how strong an indicator that word is of belonging to the category. The presence of words such as Financial, Financial Service, Credit, Loan, Payment in a company’s description is strong indicators of belonging to the financial inclusion category.

上方的“云”一词向我们展示了金融普惠公司用来形容自己的词语。 单词的大小对应于该单词属于该类别的指示器的强度。 公司描述中出现的词(例如，金融，金融服务，信贷，贷款，付款)是属于金融普惠类的有力指标。

Using the descriptions of Financial Inclusion applicants, we then trained a classifier model that would take the text description of a company as input and return a yes if the model thought it was a Financial Inclusion company or otherwise, return a no.

然后，使用“金融普惠”申请人的描述来训练分类器模型，该模型将以公司的文字描述作为输入，如果模型认为它是“金融普惠”公司，则返回“是”，否则返回“否”。

Paylater did not apply to the IIC but was classified under Financial Inclusion. We can observe a high degree of overlap between the words in Paylater’s description and the words in the world cloud above.

Paylater不适用于IIC，但归类为“金融普惠”。 我们可以观察到Paylater的描述中的单词与上面的世界云中的单词高度重叠。

We built 4 binary classifier models — one for each award category. We used Crunchbase as our global database of innovative companies and applied our models to the 410k+ companies that had English descriptions. Using our models, we identified 5% of the Crunchbase companies as addressing problems in the 4 challenge areas. Of this 5%, 33% were identified as addressing problems in financial inclusion.

我们建立了4个二元分类器模型-每个奖项类别一个。 我们使用Crunchbase作为创新公司的全球数据库，并将我们的模型应用于有英文描述的410k +公司。 使用我们的模型，我们确定了5％的Crunchbase公司正在解决4个挑战领域中的问题。 在这5％中，有33％被确定为解决金融普惠性问题。

While there appears to be significantly more activity in North America compared to the other regions, we should note that this could be a result of only looking at the English entries in Crunchbase.

尽管与其他地区相比，北美的活动似乎要多得多，但我们应该注意，这可能是仅查看Crunchbase中英文条目的结果。

The overrepresentation of North American countries is likely a result of excluding companies with a non-English description. It is also possible that Crunchbase is not the best source for companies in other regions.

北美国家代表人数过多可能是由于排除了非英语描述的公司。 Crunchbase也可能不是其他地区公司的最佳来源。

As we wanted to be conservative in our approach, the below graph only includes companies whose predict_proba values were greater than 0.75. We identified a total of 20,665 inclusive innovators in Crunchbase. 6831 of these innovators were identified as building solutions within financial inclusion.

由于我们希望采用保守的方法，因此下图仅包含其predict_proba值大于0.75的公司。 我们在Crunchbase总共确定了20,665名包容性创新者。 这些创新者中有6831个被确定为金融普惠制的解决方案。

Adding this layer of companies identified as Financial Inclusion to our map results in the below. As we did not have the data to capture the markets being served by these companies, we are only able to represent where companies are based, not necessarily where their solutions are being applied.

将这层被确定为“金融普惠”的公司添加到我们的地图中，结果如下。 由于我们没有数据来捕获这些公司所服务的市场，因此我们只能代表公司所在的位置，而不必代表其解决方案的应用位置。

结论与进一步研究 (Conclusion and Further Research)

While it is difficult to paint a complete picture, we can start to get a sense of where there may be unmet demand — this being the light blue countries with the fewest dots. These regions could be potential markets for inclusive innovation and we now start to think about how to incentivize inclusive innovation in these areas. We can also begin to look at the regions that are performing well to see what can be learned from the underlying structures or systems in place.

尽管很难绘制出完整的图片，但我们可以开始了解可能存在需求缺口的地方-这是点数最少的浅蓝色国家。 这些地区可能是包容性创新的潜在市场，我们现在开始考虑如何在这些领域激励包容性创新。 我们还可以开始查看表现良好的区域，以了解可以从现有的基础结构或系统中学到什么。

While this gives us a starting point for understanding where there is an unmet need for inclusive innovation, this research can be improved by (1) gaining access to data that can help us better understand which markets companies serve, (2) including company data on regions that may be underrepresented in Crunchbase and (3) expanding our training dataset to include both inclusive innovators and other companies to give us a sharper contrast of how inclusive innovators describe themselves.

虽然这为我们提供了一个了解哪里有包容性创新未满足需求的起点，但可以通过以下方法来改进此研究：(1)获得可以帮助我们更好地了解公司服务于哪些市场的数据；(2)包括有关以下方面的公司数据：在Crunchbase中代表性不足的地区，以及(3)将我们的培训数据集扩展到包容性创新者和其他公司，以使我们与包容性创新者如何形容自己形成鲜明对比。

This work was led in collaboration with Asa Motha-Pollock.

这项工作是与Asa Motha-Pollock合作完成的。