A New Way to Understand Listening Campaign Results

In October 2023, our Long-term Planning Committee commissioned live and virtual events. They were part of a statewide listening campaign with a goal to find out the issue and problems that Michigan residents would like to see solved over the next ten years. We held a series of live events and distributed this survey via email and social media. The primary question took people down a five-question path to telling us what was their most important thing to see changed and how it impacts their lives. 

What we will do in this piece is to use the data from a natural language understanding (NLU) analysis of survey responses to create an emotion-driven communications strategy. NLU can read text for emotion, sentiment, context and intent. 

We do this to attract new people to get involved with the Michigan People's Campaign. Someone who chooses to take 2-5 hours of their time to spend with us is the kind of person we're looking for. Our objective is to increase the number of one-on-one conversations with people interested in becoming local activists. 

The Michigan People's Campaign teaches people how to become organizers, those who use their energy to enlist others to take action around issues. One-on-ones are key to vetting a potential new activist and understanding where they can make the most impact. They are the "new blood" for the Michigan People's Campaign. The number is a critical organizational KPI.

The online survey received 54 responses, seven male and 47 female. We asked them the following questions in order:

1. What would you like to see change in your community in ten years?

2. Of those, what are the top three?

3. Why those three?

4. Which one is the most ambitious and possible?

5. How have you been impacted by your top issue?

We will dig deep into the first question, “What would you like to see change in your community in ten years?” The question was designed to give respondents an opportunity to open up and provide a longer response.

We used IBM Watson NLU, an artificial intelligence tool to analyze the text of their answers. Watson NLU  will help us to accomplish our first goal; which is to understand the motivations and context behind the answers people provided us. Watson NLU is powered by natural language understanding, a branch of artificial intelligence. The algorithm can parse text data for emotion, sentiment, syntax, semantic roles and other linguistic categories. In the next section, we will look more at Watson NLU and the dataset itself.

Model

The dataset below is an analysis of responses by keyword and emotion. You can see how the keywords fit into the text of their response, as well as the associated emotions and sentiments. 

This is one way NLU separates itself from traditional survey analysis. Generally, text answers are reviewed and used for supporting opinions. Now they add color and nuance. Generally, text responses aren't broken down by keyword. They definitely aren't analyzed for emotion and sentiment. You should be able to see the beginnings of how to create audience profiles not only based on gender, but emotion.

To fully interact with the dataset below, tap the icon at upper right to open in a new tab.

Methodology

We exported our answers from an EveryAction database hosting our survey. From there we used a Jupyter Notebook running IBM Watson NLU to perform the analysis. The processed data output itself into a Mongo database. From there we were able to start understanding the data using visualizations, tables, charts, etc. We used a ChatGPT large language model (LLM) for deeper linguistic data analysis. An LLM is a very, very, very robust chatbot trained on a single topic and dataset. We also used the tool to help with copy foundation, draft ideas, proofing and editing.

Tap here to try the LLM for yourself. Ask questions about the data, respondents and more. 

(Requires ChatGPT Plus)

Model Training Language

"You are an expert with natural language understanding; especially IBM Watson NLU. You are also a digital communication strategist, and a digital advertising  expert with a specialty in Google, Facebook and Instagram ads. You are a technology writer and data storyteller that specializes in visualizations."

Privacy and Security

To protect the identity of respondents, we changed their names and removed identifiable information from story responses. To protect the integrity of our voter file management system, we created a fictitious system of ID codes. When Watson NLU analyzes our data, we instruct it to analyze the text response; not the entire dataset. This protects the rest of the data in the set.

Understanding Scores and Emotions

Most everything in these reports are scored from 0 to 1, meaning .25 is low and .99 is extremely high. Sentiment is scored on a scale of -1 to 1, meaning zero is when a person neither likes or dislikes something. Usually referred to as neutral opinion. Emotions are defined below.

• Anger A strong feeling of annoyance, displeasure, or hostility. In NLU, anger is typically detected in language that expresses irritation, frustration, or rage.
  
  

• Disgust A feeling of revulsion or profound disapproval aroused by something unpleasant or offensive. In NLU, disgust can be identified in expressions of aversion or strong disapproval.
  
  

• Fear  An unpleasant emotion caused by the belief that someone or something is dangerous, likely to cause pain, or a threat. In NLU, fear is detected in language expressing anxiety, apprehension, or concern about potential harm or danger.
  
  

• Joy  A feeling of great pleasure and happiness. In NLU, joy is identified in language that expresses happiness, contentment, or positive excitement.
  
  

• Sadness  A feeling of sorrow or unhappiness. In NLU, sadness is recognized in language expressing grief, sorrow, or a sense of loss.

The disgust emotion looks like the hardest to manage in an audience. Anger and sadness pass. Disgust is lasting.

Confidence, Relevance and Threshold Scores

Confidence and relevance are two key metrics that confirm data accuracy. Confidence scores in natural language understanding refer to the level of certainty or reliability assigned to a particular model's interpretation or prediction of text. When a language model processes input, it generates output and an associated confidence score to indicate certainty.

For example, if you ask a chatbot a question, it may provide an answer along with a confidence score. A high confidence score suggests that the model is more confident in its response, while a lower score indicates a higher degree of uncertainty.

Relevance refers to how closely the output or response generated by a model aligns with the input or the user's intent. It's a measure of how well the model understands and addresses the user's query or statement.

There is a relationship between relevance and confidence scores. High confidence scores are often used to quantify the model's certainty of its understanding. A highly relevant response is typically associated with a high confidence score, indicating that the model is confident in its interpretation and the subsequent output.

For example, if you ask a language model about the weather in a specific location, a relevant and confident response would provide an accurate current forecast for the locality.. 

We can also use thresholds to address shortcomings in the data. In our case, 54 responses is likely just enough responses to build a strategy around.  We say that because sometimes accuracy is affected by simply not having enough data. From our experience, when there is too little data you see an overwhelming number of duplicates in the categories and concepts area. Intent is an area to pay close attention to as well. Use only the clearest and most direct connections. 

Setting thresholds in natural language understanding involves defining criteria for confidence scores or other metrics to filter or control the model's output. For example, you might You can set a confidence or relevance threshold to filter out responses with low scores. This helps ensure that only highly confident and presumably more accurate answers are presented to users.

Audience Personas

Each of the personas below were created based on specific data points from our dataset above. We asked ChatGPT to take on the persona of a digital marketing strategist, an expert on natural language understanding and a communicator. ChatGPT reviewed eight different parameters, from emotion to semantic roles in order to create narratives and profiles that accurately reflect the interests and motivations of respondents. Once established, these profiles can guide the targeting and messaging strategies, ensuring that the content resonates with and are relevant to each audience.

Emotional Analysis

Emotion in Natural Language Understanding (NLU) refers to the ability of a system to recognize, interpret, and process the emotional content in text.

In the interactive visualization above, you see that respondents are generally positive and even tempered about their visions of the future. Emotional scores tend to be slighly below moderate at between .3 and .5. 

Our read of the visualization says there are many things that would make our respondents happy to see accomplished. At the same time they feel especially deeply about those things that make them sad. They are even-tempered about their visions. The sentiment trend line is largely flat, pointing slightly upwards.

Keywords

Keywords are specific words or phrases that hold significant meaning within a text. In NLU, these are the terms that are identified as crucial for understanding the content and context of a conversation or document.

Emotional Combinations in Responses

Humans respond to things with a mix of feelings at the same time. A man sees his son score in the big game has feelings of pride and sadness that his own father didn't live long enough to see it.  Looking at secondary emotions adds nuance to what's on people's hearts and minds. The analysis of the most prevalent combinations of dominant emotion and the second highest emotion in the responses reveals the following top combinations:

• Sadness (Dominant) and Joy (Second Highest): This combination occurred 201 times.

• Joy (Dominant) and Sadness (Second Highest): This combination was observed 194 times.

• Sadness (Dominant) and Disgust (Second Highest): Occurred 138 times.

• Sadness (Dominant) and Sadness (Second Highest): Recorded 101 times.

• Sadness (Dominant) and Fear (Second Highest): Found in 58 cases.

It is worth taking a closer look at each combination, but especially sadness/sadness and sadness/fear combinations. Are there clear inverse relationships, based on responses?  If so, can we create a clear and compelling message from it?

Top Sadness-Joy Keyword Combinations

• Force, Affordable housing, abortion rights, best case scenario wishlist, competition.

Top Joy-Sadness Keyword Combinations

• Resources, public schools, children, Stronger programs, accessible community-based healthcare.

Top Sadness-Disgust Keyword Combinations

• First day coverage, employer, sustainable rates, style health insurance, national availability.

Top Sadness-Fear Keyword Combinations

• Better training, resources, smaller school districts, special Ed, school employee turnover - however.

Next, let's look at keywords by gender.

Granted, seven of 54 responses from men make it difficult to make assumptions. However, it would seem prudent to target campaigns to an audience heavy with women.

Another place to find elow is a visualization of the subjects of their responses. Again, an analysis that a traditional survey usually doesn't provide:

Subject Text

NLU is superior to traditional survey analysis in terms of being able to pickup more details than reading and taking notes. A very capable person could do it with 54 responses. But what of their own experiences and and thoughts would they project on the data if there were 154 responses? NLU replaces supposition with data.

The broader strokes of change are becoming clearer. Let's look at another aspect of the data, concepts.

Concepts

Concepts in Natural Language Understanding (NLU) enable us to understand the context or the broader topic of the content. They are used to categorize and theme content. Concepts represent abstract ideas that are often broader than specific instances or entities and help in grasping the meaning of text beyond mere keywords. 

Here's an easy example. If a person writes, "cruller," "glazed," and "chocolate" to describe their favorite donuts, you should expect some of the related concepts to be "breakfast," "snacks," and "food." Concepts allow you to more easily see how to classify keywords.

Looking at the visualization above, the first thing that jumps out to me are the terms “affordable housing,” “health care” “better schools,” “homelessness,” and “health insurance.” They are rich with relevance and emotion, and are by far what people want to see in future change. They also look like a ready-made list for a political agenda. In general, the array matches the balanced interests of respondents.

The disgust concepts also stand out, not just because they're in pink. Knowing how debilitating disgust can be sparks curiosity at the stories behind the terms. Perhaps some of these stories could be used for campaigns or ad material. As stated earlier, anger and sadness pass. Disgust lingers. If you can find a way to effectively tap into that emotion, you've done something.

The important thing is NLU provides much more accurate insight behind something humans usually project their own emotions and experiences upon: rows and columns of data.

Sentiment

Sentiment refers to the identification and categorization of polarity  (positive, neutral, negative) in opinions or feelings expressed in a piece of text. 

Document Sentiment

The analysis of the overall sentiment in the dataset reveals the following:

• Negative Sentiment: Accounts for approximately 50.13% of the documents.

• Positive Sentiment: Represents about 48.81% of the documents.

• Neutral Sentiment: Makes up a minor portion, approximately 1.07% of the documents.

Document Sentiment Trends and Patterns

Predominance of Polarized Sentiments: The dataset is characterized by a high prevalence of polarized sentiments (either positive or negative), with very few instances of neutral sentiment. This suggests that the content being analyzed tends to elicit strong opinions or reactions, either positively or negatively.

Near Balance Between Positive and Negative Sentiments: The distribution between positive and negative sentiments is almost equal, indicating a diverse range of opinions or perspectives within the documents.

Limited Neutral Perspectives: The notably small percentage of neutral sentiments could indicate that the topics or contents of these documents are such that they inherently invoke a clear stance, opinion, or feeling, leaving little room for neutral or indifferent reactions.

Entity Sentiment

Entity sentiment in the context of Natural Language Understanding (NLU) refers to the sentiment specifically associated with identified entities in a given text. Entities are typically names of people, organizations, locations, or other significant nouns. In entity sentiment analysis, the goal is to determine not just the overall sentiment of the entire text but the specific sentiment directed towards each identified entity.

The analysis of entity sentiment in the dataset reveals the following:

• Positive Sentiment represents about 56.40% of the entities.

• Negative Sentiment accounts for approximately 25.57% of the entities.

• Neutral Sentiment makes up around 18.03% of the entities.

Entity Sentiment Trends and Patterns

Dominance of Positive Sentiment: A significant majority of entities are associated with positive sentiments. This indicates that, on balance, the entities discussed or mentioned in the dataset are viewed more favorably.

Negative Sentiment Is Substantial but Less Prevalent: Negative sentiments are also a notable part of the dataset but are less common than positive sentiments. This suggests a certain level of critical or unfavorable views towards some entities, albeit less dominant than positive views.

Higher Presence of Neutral Sentiment Compared to Document Sentiment: Unlike the overall document sentiment, the neutral sentiment for entities is more pronounced. This could mean that while the documents themselves tend to provoke stronger sentiments, the specific entities mentioned within them might not always elicit such strong reactions.

Polarization Is Less Pronounced: Compared to the overall document sentiment, the polarization in entity sentiment is less stark. There's a more balanced distribution across positive, negative, and neutral sentiments. When we say that polarization is less pronounced in entity sentiment compared to the overall document sentiment, it means that the sentiments associated with specific entities (individuals, organizations, topics, etc.) within the documents are more evenly distributed across positive, negative, and neutral categories.

Categories

Categorization in NLU is widely used for organizing and filtering large volumes of text data, such as in content management systems, for enhancing search and discovery in digital libraries, or in business intelligence for analyzing customer feedback and market trends. It helps in quickly identifying the relevant texts for further analysis or action.

Categories

The category analysis of the dataset reveals the following:

Most Common Categories

• Education-Related Categories: The top categories are heavily skewed towards education, with specific focus areas like:

  • Studying Business (4.98%)

  • Special Education (4.15%)

  • Teaching and Classroom Resources (4.15%)

  • Homework and Study Tips (3.71%)

• Healthcare: Another significant category is healthcare, particularly in the context of welfare, accounting for 3.30%.
  
  

Diverse Range of Categories

The dataset includes a wide variety of categories, totaling 102 unique ones, indicating a diverse range of interests and concerns among the respondents.

Trends and Insights

• Strong Focus on Education and Learning: The predominance of education-related categories suggests that respondents have a keen interest in the educational sector. This could reflect a societal emphasis on educational reform or improvement.

• Healthcare as a Key Concern: The prominence of healthcare in societal welfare indicates its importance in public discourse, possibly reflecting concerns about healthcare accessibility, quality, or reforms.

• Broad Range of Interests: The presence of numerous categories, some with very few mentions, highlights a broad array of interests and concerns, from business travel to aging and fitness.

Relations

Relations data refers to the identification and analysis of the relationships between entities within a text. This aspect of NLU involves understanding how different entities (like people, organizations, locations, etc.) are connected to each other based on the context provided in the text.

Entity Relationships: Relations data primarily involves understanding how different entities (like people, places, organizations, dates, etc.) in a text are related to each other. For example, in the sentence "Alice works at Acme Corp.," the relationship between "Alice" (a person) and "Acme Corp." (an organization) is that of employee-employer.

Contextual Understanding: It goes beyond merely recognizing entities, focusing on the context and the nature of their interactions or connections. This might involve identifying whether an entity is the subject, object, beneficiary, cause, or location in relation to another entity.

Types of Relationships: Common types of relationships include causality, attribution, part-whole, membership, location-based, and temporal relationships.

Top relation types, arguments and contexts

Top Relation Types

1. locatedAt: 36 occurrences

2. employedBy: 28 occurrences

3. residesIn: 20 occurrences

4. partOfMany: 16 occurrences

5. basedIn: 16 occurrences

Top Arguments

1. schools: Mentioned 10 times

2. communities: Mentioned 8 times

3. people: Mentioned 6 times

4. assault: Mentioned 6 times

5. our: Mentioned 6 times

Contexts of Top Relations

1. agentOf: Related to improving voting opportunities and election day processes.

2. basedIn: Associated with public transportation, affordable health care, and environmental justice.

3. educatedAt: Connected to better security at schools.

4. employedBy: Pertains to free Medicare health insurance for people who cannot afford it.

5. instrumentOf: Linked to affordable health care for all and sustainable safety measures.

Entity Analysis

Entities are distinct elements within a text that represent real-world objects, concepts, or individuals. These entities are typically nouns or noun phrases and are integral for extracting meaning and context from language data.

Top Entities

• Teachers: Most frequently mentioned entity, indicating a significant focus or concern related to this group.

• Detroit and Michigan: These geographical entities suggest a regional focus in the data.

• Numeric Entities ("1" and "2"): Their inclusion could be indicative of ranking, prioritization, or categorization within the context.

Sentiment Analysis

• Negative Sentiment: Entities associated with a negative sentiment scored around -0.698 on average, indicating strong negative perceptions or associations.

• Neutral Sentiment: Neutral sentiment maintains a score of 0.0, suggesting a balanced or unbiased view towards certain entities.

• Positive Sentiment: Positive sentiment averages around 0.769, showing a generally favorable view towards some entities.

Emotional Context

• Anger: Averaging at 0.052, this score is relatively low, indicating that the entities do not generally evoke strong anger.

• Disgust: With an average score of 0.102, there is a slightly elevated sense of disgust associated with some entities.

• Fear: The average fear score is 0.066, suggesting a moderate level of fear or concern related to the entities.

• Joy: Joy scores average at 0.228, implying a significant presence of positive, joyful associations.

• Sadness: The highest average emotional score at 0.277, indicating that sadness is a prominent emotion in relation to the entities.

Entity-Specific Insights

• Detroit and Michigan: These entities show positive sentiment, with Detroit notably eliciting high joy scores. This could reflect positive attitudes or hopeful perspectives related to these locations.

• Teachers: This entity has a balanced mix of emotions but notably high sadness scores. This might reflect complex perceptions or challenging issues related to the teaching profession.

• Numeric Entities: Exhibiting negative sentiment, these might be indicative of issues or topics that are viewed unfavorably.

Subject/Verb/Object Analysis

Subject/verb/object analysis is a fundamental part of NLU, serving as a building block for more complex language understanding tasks. It provides a basic framework for parsing sentences and extracting meaning.

Subject (S): The subject is the person, thing, or entity that performs the action or is described in the sentence. It is usually a noun or a pronoun. For example, in the sentence "The cat sits on the mat," "the cat" is the subject.

Verb (V): The verb expresses the action performed by the subject or the state of being. It is a crucial part of a sentence that indicates what is happening. In the same sentence, "sits" is the verb.

Object (O): The object is the entity that is affected by the action of the verb. It often receives the action. In our example, "the mat" is the object, as it is the thing being sat on.

Understanding Intent & Verb Tense

In Natural Language Understanding (NLU), "intent" refers to the purpose or goal behind a user's input, typically in the form of text or speech. Understanding intent is a critical aspect of NLU systems because it helps determine what the user wants to achieve. For our purposes, "intent" refers to the actions respondents are willing to take in order to see their ten-year vision come true. 

If we can find clear relationships between verbs and objects, we literally see what people intend to do.

The twist that natural language understanding adds is tense. The infinitive tense is key for advocacy groups because it is usually preceded by a version of "to". In the visualization above, the verbs in black are infinitive. So "to Eliminate", "to combat," etc. It is a purpose-filled term because it is not linked to a timeline or status. Infinitive verbs in this context represent the willingness to fight onward for whatever the object of these verbs are.

Common Verb Tenses

Each tense conveys a different time frame and aspect (the nature of the action). Understanding these nuances is key in language learning, communication, and in the field of NLU for processing and generating language accurately.

Simple Present

• Use: Describes habits, general truths, and repeated actions. It's also used for fixed arrangements, or schedules.

• Example: "She writes a blog."

Simple Past

• Use: Describes actions completed in the past. It's used for actions that happened at a specific time, which is either mentioned or understood.

• Example: "She wrote a blog last year as well."

Simple Future

• Use: Describes actions that will happen in the future. It's often used with 'will' or 'shall.'

• Example: "She will start a blog next year."

Past Participle

• Use: Forms perfect tenses and the passive voice. It's the form of the verb that is used after "have" in perfect tenses or after "be" in passive constructions.

• Example: "The blog was written by her." (Passive), "She has written all of the blog posts." (Present Perfect)

Now let's look at objects. Objects are the word or phrase that receives the action of the verb. Looking at the visualization below, you can easily imagine what the preceding verbs could be.

Objects

Emotion Analysis, by Object

By analyzing the emotion scores and sentiment analysis across different keywords (objects), this report offers a nuanced understanding of how various topics are emotionally perceived by respondents. Each keyword's emotion profile reflects the complexities of public sentiment on these issues.

Dominant Emotion Analysis

• Example: The keyword "Medicare for All" has been identified with a dominant emotion of sadness, with a score of approximately 0.235. This suggests that when respondents mention "Medicare for All," it is often in a context that evokes sadness.

• Interpretation: The prominence of sadness in relation to this keyword could imply concerns or frustrations regarding healthcare policy, reflecting a sentiment of disappointment or dissatisfaction among the respondents.

Emotion Scores for Each Keyword

• Example: Looking at the keyword "school boards," we see it also has scores for other emotions like anger (approximately 0.052), disgust (approximately 0.226), and fear (approximately 0.129), besides sadness.

• Interpretation: This distribution of emotion scores indicates a complex emotional response towards school boards, possibly hinting at contentious or divisive issues related to education policies or administration.

Sentiment Analysis

• Example: Alongside the emotion scores, sentiment analysis reveals that both "Medicare for All" and "school boards" are associated with a negative sentiment, with a sentiment score of around -0.657.

• Interpretation: The negative sentiment score aligns with the high sadness emotion score, reinforcing the notion that discussions around these topics are predominantly negative.

Keyword-Specific Emotion Scores

• Example: For "Medicare for All," the emotion scores specifically linked to the keyword are: Anger (0.052), Disgust (0.226), Fear (0.129), Joy (0.012), and Sadness (0.235).

• Interpretation: The highest score is for sadness, followed by disgust and fear, suggesting a range of negative emotions associated with this topic. The low joy score indicates minimal positive association.

Comparative Analysis Across Keywords

• Example: If we compare the emotions linked with "Medicare for All" to another keyword like "Dems" (Democrats), we might see different emotion distributions, providing insights into how different objects/topics are perceived emotionally.

Contextual Analysis of Objects

The Contextual Analysis of Objects provides a deep dive into how specific objects are situated within the broader narrative of the respondents' views, reflecting complex interactions between subjects, objects, and verbs in the text. This analysis helps in understanding not just what objects are being discussed, but also the broader narrative and context in which they are placed. ​

Subject-Object-Verb Relationship Analysis

• Example: In one response, the subject "the state government" is associated with the verb "finance" and the object "the inequity."

• Interpretation: This implies that the state government is seen in the context of addressing or being related to inequity, specifically through financial means.

Object Keyword Contextualization

• Example: The object "deep ties to Zionists and China - reach across aisles" is linked with the subject "companies" and the verb "have."

• Interpretation: This suggests a perception that companies have complex international and political connections, as indicated by the phrase "deep ties to Zionists and China." The part "reach across aisles" could imply a call for bipartisan cooperation or broader political engagement.

Verb Tense and Contextual Meaning

• Example: The use of the present participle tense in "the state government financing the inequity" indicates an ongoing action or perception.

• Interpretation: This tense usage can imply that the issue of inequity is a current and continuing concern in the context of government financial policies.

Comparative Analysis of Object Contexts

• Example: By comparing different object keywords like "inequity" and "Zionists," we can see how different objects are contextualized in varied ways – inequity in the context of government action, and Zionists in the context of corporate ties.

• Interpretation: This comparison helps in understanding the diverse ways objects are perceived and discussed within different societal and political contexts.

Overall Contextual Themes

• Insight: The report reveals underlying themes and concerns among respondents, such as government responsibility, corporate influence, and international relations, through the lens of how objects (keywords) are discussed.

Campaign Strategy

We will use the data to build a proposed emotionally-informed communications strategy. Since our respondents are politically active in their communities, it makes perfect sense to attract more people like them to get involved with the Michigan People's Campaign. This means we will build our strategy around emotion and shared opinion instead of demographics. 

We have profiles and narratives to build a content campaign around. We have a treasure trove of data on what they think and how they feel about the issues they selected. We can use that to make our digital ads more effective. Lastly, we use the data to build SEO campaigns and landing pages that find and move people more quickly down the commitment funnel (instead of a marketing funnel).

Our objective is to increase the number of one-on-one conversations with people interested in becoming activists. The Michigan People's Campaign teaches activists how to become organizers, or people who use their energy to organize others around issues. 

One-on-one conversations are key to vetting a potential new activist and understanding where they can make the most impact. Since they measure the amount of  "new blood" coming into the Michigan People's Campaign, the number of one-on-ones is a critical organizational KPI.

Campaign Target Audience

How our audience "feels" and their intent

The traditional target audience for an advocacy group like the MPC is a mix of the usual demographics and behaviors. "Liberal," and "politically motivated" are the easy ones. But the fact that we've got natural language understanding data means we can do something completely unique: target at an emotional level. 

Understanding emotion, sentiment and word usage means we're looking to target people that feel like our respondents; no matter demographics or economic standing. That deeper connection means a lot. 

The following charts represent our target audience. These are the things they care about:

Using relevance as one of the visualization's main metrics grounds it in reality and believability. MPC is primarily looking for people motivated by the largest keywords--people, affordable housing, affordable health care, better schools and education. The remaining keywords matter but it's easier to see which are less relevant. I'd also like to take a moment to remind that a traditional survey will never reveal emotion without lots of extra questions.

NLU makes framing the emotion easier and much more accurate. Humans look at the numbers in survey data and frame their understanding through a lens of their own experience.  If the keywords in the visualization above were in black, we'd use a mix of OUR feelings and the relevance scores to attribute emotion. NLU provides the missing emotional data in the results. 

But "joy" as we know it doesn't easily translate to "education." Save for getting into the university we really wanted, we don't get "joyful" over education. Which points to where humans come in: we need to frame or adjust the focus of the related emotion. In this case, "joy" is likely closer to optimism than happiness. Optimism about better education is a better frame for thinking about the future of a person or a society. 

I'd say solemnity is the frame for sadness. Some keywords show that people are sad about their situation. Other keywords can be attributed to empathizing with the sadness of others. Marshaling a feeling of determination about the issues behind these keywords can be a way to get these people motivated. 

You can also see determination in the intent of their objectives in the visualization below.

Now let's add a layer of intent to how they feel.

The value of analyzing verbs or action words is in determining intent. Meaning, if you can find indicators whether a person prefers. one option over another, you have an ability to prepare a response for each option.  For example, if you know a person already accepts your points and arguments about a sports team, the two of you can get right to the point. It's a completely different conversation if you know the person is unaware of your team. 

Below is a chart of the top 25 subject/verb/object combinations:

With 18 combinations, it's worth looking at the "employer/sponsor/first day coverage" combination. It's one of those things that kinda sounds right sitting at the top of the data. 

Sometimes it's that easy. Most times it's not.

Strategy  I: Content

Topic Modeling

We performed an LDA (Latent Dirichlet Allocation) topic analysis using the document keywords data. An LDA is a type of statistical model that is used to classify text in a document to a particular topic. It helps in identifying which words are closely related in terms of context and meaning, thereby grouping them into topics to use for content creation.

The LDA topic analysis on our response data shows five main topics. Here are the top 10 words in each topic, which give us an idea of the themes:

Topic 0 

• Theme: Focuses on public health and safety, community welfare, and affordable living essentials.

• Keywords: care, health, public, communities, safe, income, schools, affordable, water, home

Topic 1 

• Theme: Emphasizes support for rights and resources, with a focus on children, energy, and infrastructure.

• Keywords: support, rights, Michigan, energy, zones, children, resources, green, new, buildings

Topic 2 

• Theme: Centers around healthcare, education, justice, and control measures in various sectors.

• Keywords: better, healthcare, schools, debt, gun, food, medical, justice, universal, control

Topic 3

• Theme: Related to education and school systems, including teacher-related aspects, reform, and growth.

• Keywords: education, people, school, teacher, reform, programs, insurance, justice, attrition, growth

Topic 4 

• Theme: Focuses on housing, community services, education, and environmental concerns.

• Keywords: housing, affordable, community, medicare, trees, teachers, services, crime, students, fossil fuels

Channel Strategy

Social Media

• Thematic Posts: Regular posts focusing on themes like healthcare advocacy, educational reform, and community development. Use compelling visuals and engaging stories.

• Engagement Campaigns: Interactive campaigns, like Q&A sessions, live discussions, and polls about social issues, to increase engagement and awareness.

• User-Generated Content: Encourage volunteers to share their experiences and stories, creating a sense of community and authenticity.

Print Ads

• Inspirational Stories: Feature real stories of volunteers making an impact in areas like healthcare and education. Use high-impact visuals and compelling narratives.

• Informational Ads: Highlight the need for volunteers in different sectors, backed by statistics and facts derived from your data.

Community Radio

• Interview Segments: Air interviews with volunteers and leaders in your organization, discussing various initiatives and the impact of volunteer work.

• Public Service Announcements: Create short, impactful messages about the importance of volunteering, tailored to resonate with different audience segments.

Video Channels and Community CATV

• Documentary-Style Features: Develop short documentaries showcasing the work and impact of your volunteers in various fields.

• Advertisement Spots: Create engaging and emotionally compelling ads that highlight the need for volunteers and the difference they can make.

Blog Posts

• Feature Stories: Write detailed stories about individual volunteers or specific projects, showcasing the human side of volunteering.

• Educational Content: Publish articles that educate readers about issues in healthcare, education, and community development, and how volunteering helps.

Feature Stories (for Online and Print Media)

• Volunteer Spotlights: Deep dives into the lives of volunteers, their motivations, and the impact they have made.

• Behind-the-Scenes Look: Share insights into how your organization operates, the planning of initiatives, and the execution of projects.

General Content Guidelines

• Consistent Messaging: Ensure that the core message of making a meaningful impact through volunteering is consistently conveyed across all channels.

• Target Audience: Tailor the content to resonate with the identified audience profiles – from socially conscious individuals to policy and governance enthusiasts.

• Visuals and Tone: Use compelling visuals and maintain a tone that is inspiring, informative, and empathetic.

• Call-to-Action: Each piece of content should have a clear CTA, encouraging the audience to learn more, sign up to volunteer, or engage with your organization in some way.

• Cross-Promotion: Use each channel to promote content on the others, creating a cohesive and integrated marketing effort.

This content strategy leverages the insights from the data to create targeted, engaging, and impactful content across various channels, effectively communicating the need for volunteers and encouraging audience participation.

Strategy  II: Google Ads

Target Audience Insights

• From the sentiment analysis, it appears that health care, education, and government-related topics are significant. This implies your audience might be interested in social issues, healthcare, and governance.

• Emotions like sadness and concern about topics like "Medicare for All" and "school boards" suggest targeting individuals interested in social welfare and education reform.

Keyword Strategy

• Keywords like "community", "Medicare for All", "Dems", "Help", and "school boards" are relevant. Use these for targeting and in your ad copy to resonate with your audience's interests.

• Since "health care" and "education" are prominent, consider including them in your keyword strategy.

Ad Copy and Creative

• Focus on emotional appeals that align with the sentiments and concerns found in the data. Highlight how volunteering can impact healthcare, education, and community welfare.

• Use phrases that emerged from the sentiment analysis, like "turning visions into reality", to create a compelling narrative.

Landing Page Optimization

• Ensure your landing page reflects the themes and emotions identified in the data. Include stories or testimonials about the impact of volunteering in relevant fields.

• Make the call-to-action (CTA) clear and compelling, focusing on the difference volunteers can make.

Ad Formats and Placement

• Considering the audience's interest in social issues, use video ads to tell compelling stories. Platforms like YouTube can be effective for this.

• Utilize responsive search ads with multiple headlines and descriptions to test which messages resonate the most.

Campaign Structure and Targeting

• Create separate ad groups for different themes (e.g., healthcare, education) to better tailor the ad copy and keywords.

• Use demographic and interest-based targeting to reach individuals likely interested in social welfare, education, and community involvement.

Performance Tracking and Optimization

• Regularly review the campaign's performance, focusing on metrics like click-through rate (CTR) and conversion rate.

• Continuously refine your keywords, ad copy, and targeting based on performance data.

• These suggestions are based on the insights from the provided data. Tailoring your campaign to align with your audience's sentiments and interests should help in achieving your goal of increasing clickthroughs to the recruitment landing page. 

Proposed Ad Groups

Healthcare Advocacy

• Focus on keywords and ad copy related to healthcare issues, reforms, and advocacy. This can resonate with individuals interested in healthcare as a social issue.

Educational Initiative

• Target those interested in education reform and involvement in school boards. Use keywords and messages related to education, community involvement in schools, and educational equity.

Community Development

• Create ads centered around community improvement and engagement. This can attract individuals interested in local community work and social welfare.

Social Welfare and Policy

• Focus on broader social welfare topics and policies, targeting individuals interested in social justice, government policies, and societal improvements.

Volunteer Engagement

• Specifically target those looking to volunteer, using keywords like "volunteer opportunities", "community service", and "make a difference".

Youth and Student Involvement

• If your data shows interest from younger demographics, consider a group targeting students and young adults interested in activism and volunteerism.

Each ad group should have tailored ad copy that speaks directly to the interests and concerns of the audience segment it's targeting. For instance, ads in the Healthcare Advocacy group could highlight the impact of volunteers in health-related initiatives, while those in the Educational Initiatives group could focus on the role of volunteers in shaping educational policies and supporting schools.

Proposed Ad Types

Responsive Search Ads and Samples

Responsive search ads allow you to enter multiple headlines and descriptions, and Google automatically tests different combinations to learn which performs best. This is ideal for honing in on the most effective messaging for your audience. Tailoring headlines and descriptions to include keywords and themes such as "volunteer opportunities", "community development", "healthcare advocacy", or "education reform" can resonate with the interests identified in the data.

Ad Group: Healthcare Advocacy

• Headlines: "Join Our Healthcare Volunteer Team", "Make a Difference in Healthcare", "Volunteer for Health Advocacy"

• Descriptions: "Be a part of meaningful healthcare change. Volunteer today to make a real impact in our community's health services.", "Looking for a way to contribute to healthcare reform? Your volunteer efforts can make all the difference."

Ad Group: Educational Initiatives

• Headlines: "Volunteer in Education Reform", "Help Shape the Future of Education", "Get Involved in School Boards"

• Descriptions: "Your passion for education can help shape our future. Join us as a volunteer in educational initiatives.", "Be the change in our educational system. Volunteer your skills and time to make a lasting impact."

Ad Group: Community Development

• Headlines: "Community Development Volunteers Needed", "Build a Better Community with Us", "Your Community Needs You"

• Descriptions: "Join our team of volunteers dedicated to community development and make a tangible difference where it matters most.", "Contribute to your community's growth. Volunteer in local projects and initiatives that drive change."

Video Ads & Samples

Video ads, particularly on platforms like YouTube, can be powerful for storytelling. They can visually and emotionally engage the audience, illustrating the impact of volunteering and the difference they can make in areas like healthcare, education, and community welfare. These ads can be targeted based on user interests, search history, and demographics, making them a strong tool for reaching a receptive audience.

Healthcare Advocacy

• A touching video showcasing volunteers at a healthcare event, with testimonials on how their work has impacted the community. The call to action (CTA) invites viewers to join the cause.

Educational Initiatives

• A dynamic video featuring scenes from educational programs and school board meetings, highlighting the need for volunteer involvement. The CTA encourages viewers to get involved in shaping education.

Community Development

• A video tour of various community development projects, with volunteers sharing their experiences and encouraging viewers to join and make a difference in their own neighborhoods.

Display Ads & Samples

Display ads are effective for reaching a wider audience across various websites and apps. They can be used to increase brand awareness and remind users of your cause. Using compelling visuals and clear calls-to-action (CTAs), these ads can effectively direct traffic to your landing page. With Google's targeting options, you can focus on users who have shown interest in related topics like community service, social welfare, and educational initiatives.

Healthcare Advocacy

• An image of volunteers vaccinating seniors, with a banner saying "MPC is at work the community, vaccinating at least 200 seniors a month. Help us do more" and a CTA button "Do More than Vote".

Educational Initiatives

• A compelling visual from a first-person story on how badly school staff managed education policies: "Shape Tomorrow's Education Today" and a CTA "Get Involved".

Community Development

• A photo collage of community projects with text overlay: "Your Community Needs You - Be a Volunteer Hero". Includes a prominent "Learn More" button.

Each ad set is designed to resonate with the target audience's interests and motivations as identified in the provided data, with a clear and compelling call to action guiding them towards your recruitment landing page.

Strategy III: SEO

For SEO purposes, effectively using the insights from the LDA analysis involves more than just tagging. Here’s a comprehensive approach:

Tagging Strategy

Utilize Top Keywords from Each Topic

• From Topic 1 (Education and Healthcare), use tags like "education resources", "healthcare tips", "classroom activities".

• For Topic 2 (Financial and Real Estate), consider tags such as "financial aid", "real estate trends", "scholarship opportunities".

• In Topic 3 (Health and Family), tags like "mental health", "family care", "parenting tips" would be relevant.

• For Topic 4 (Politics and Government), use tags like "political analysis", "government policies", "law and ethics".

• In Topic 5 (Business and Industrial), consider tags such as "business strategies", "industrial developments", "education in business".

Other SEO Considerations

Content Creation and Optimization

• Develop blog posts, articles, and web pages that focus on these topics. Ensure the content is informative, engaging, and uses the keywords naturally.

• Update existing content to include relevant keywords and phrases identified in the LDA analysis.

Meta Descriptions and Title Tags

• Craft compelling meta descriptions and title tags for each web page, incorporating the top keywords. This not only helps in SEO but also improves click-through rates from search engine results pages.

Internal Linking

• Use internal linking to guide visitors to related content on your site. This helps in keeping the audience engaged and also allows search engines to crawl your site more effectively.

Backlink Strategy

• Aim to get backlinks from reputable sites within the domains of your key topics. For example, seek links from educational or healthcare sites for Topic 1.

Local SEO (if applicable)

• If your organization operates locally, ensure to optimize for local SEO. This includes having a Google My Business listing and getting listed in local directories.

Mobile Optimization

• Ensure your website is mobile-friendly, as a significant amount of searches are conducted on mobile devices.

Voice Search Optimization

• Optimize for voice search by using more conversational keywords and phrases, as people tend to use natural language for voice queries.

Image Optimization

• Use relevant images and optimize them with descriptive file names and alt tags that include your keywords.

Regular Monitoring and Updating

• SEO is not a one-time task. Regularly monitor your website's performance using tools like Google Analytics and Google Search Console, and update your strategies accordingly.

By integrating these SEO strategies with the insights from the LDA analysis, you can significantly enhance the visibility and reach of your content in search engine results, attracting a more targeted and engaged audience.

LDA Analysis Keywords

More on Voice Search

Voice search optimization is increasingly important as more people use voice-activated devices like smartphones, smart speakers (like Amazon Echo and Google Home), and virtual assistants (such as Siri, Google Assistant, and Alexa) for internet searches. The process is focused on adapting to the way people use voice commands for searches, which differs significantly from traditional text-based queries. Here are key aspects of voice search optimization along with examples:

Natural Language and Conversational Tone

• Voice searches are typically more conversational and longer than text searches.

• Example: A text search might be "best coffee shops NYC," whereas a voice search would be "What are the best coffee shops in New York City?"

Question-Based Queries

• Many voice searches are phrased as questions. Optimizing content to answer these questions can improve visibility in voice search results.

• Example: Content could be optimized to answer questions like "How can I volunteer for healthcare initiatives?" or "What are the latest trends in educational reform?"

Local Search Optimization

• Voice searches often have a local intent. People use voice search to find nearby services or locations.

• Example: "Where is the nearest volunteer center?" or "Find a community development program near me."

Featured Snippets and Position Zero

• Voice search devices often read out the featured snippet or the top result. Therefore, ranking in this position can increase the likelihood of being the chosen voice search response.

• Example: Crafting content that directly answers questions like "What are the benefits of community volunteering?" in a concise manner can help in appearing in this prime position.

Mobile-Friendly Website

• Since many voice searches are done via mobile devices, having a mobile-optimized website is crucial.

• Example: Ensuring your website has a responsive design, fast loading times, and easy navigation on mobile devices.

Schema Markup

• Using schema markup can help search engines understand the context of your content, making it more likely to be used in voice search results.

• Example: If you're posting an event about a local educational reform seminar, use the appropriate Event schema markup to help search engines understand the details.

Long-Tail Keywords

• Voice searches often involve long-tail keywords. Including these in your content can help match with voice queries.

• Example: Instead of targeting short keywords like "volunteer opportunities", target long-tail phrases like "where to find volunteer opportunities for healthcare in Boston".

Improve Local SEO

• For businesses and organizations, optimizing for local SEO is important for voice search, especially for 'near me' queries.

• Example: Ensure your Google My Business listing is up-to-date, with accurate location, contact details, and business hours.

FAQ Pages

• Creating FAQ pages can be an effective way to incorporate question-based, conversational phrases.

• Example: An FAQ page for a community development organization might include questions like "How can I contribute to local community development?"

By optimizing for these aspects, you can enhance your content's suitability for voice search queries, improving your visibility in this increasingly popular search mode.