Nokhuthula Njoko said that she feels empowered by being involved with Mobenzi.

“It helps me focus on the work I’m doing, knowing that for each task I complete I will be paid. I’ve learnt to be a critical thinker and enjoy the challenge of learning new words and abbreviations in different languages and in business terms. I’ve even made myself a book where I write down all the words I don’t understand, then later find their meanings. Mobenzi has given me employment, empowerment and an education in the business world”.

Trevor Ngcobo said being a Mobenzi agent enables him to study and still work at his own leisure.

“I never have to worry about transport problems, being late for work or not having time to attend college. I can make money, study and even do my Mobenzi tasks in a taxi on my way to lectures. It’s helped me in more ways than I thought when I first started”.

Msizi Phewa relishes in the fact that he can tell his peers that he ‘works on the internet and analyses data’.

“It makes me feel so important when I tell people that I work with analysing information from social networking sites like Twitter and Facebook. And because it’s something you can do in your spare time and you are paid for it, your mind is not focused on distractions of drugs, alcohol and hanging out with unproductive and negative people on the streets. I want the world to eventually be plugged into Mobenzi so that we can have an entire planet of productive people”.

It was great to hear such positive and interesting feedback.

On May 26th we invited some representatives from the press to the launch of the next phase of our Mobenzi pilot project in KwaNyuswa (Valley of a thousand hills, KZN).

During our two week trial run in December 2009, agents had completed Mobenzi tasks using our company owned phones, under supervision and together at a central location. Since May 26th however, a group of agents have been working independently as private contractors to Mobenzi.

These are some of the major factors that make the launch of this phase of the pilot a significant step forward.

1. Agents are now working in their own time, requesting batches of tasks whenever they have a few minutes spare.
2. They complete tasks while at home, travelling on public transport or even between lectures at college.
3. Most of the agents are using their own mobile phones after having installed the Mobenzi application from a link we sent to them.
4. With each task that agent’s complete, associated credit is built up in their account. Once credit reaches a certain thresh-hold, funds are disbursed electronically to their phones using FNB’s SendMoney platform. Although some agents had to borrow our company phones, many have already earned enough income from Mobenzi to purchase their own, brand new compatible Nokia phones.

These changes in the way the pilot is being run are allowing us to test the scalability of the concept. We can now manage recruitment of new agents, assignment of tasks, monitoring of quality and disbursement of funds all from our central office.

With this platform in place, it is only the demand from businesses for the services of our agents that will slow the growth of Mobenzi.

Traditionally, when it comes to gathering data via SMS, researchers have two rather ungainly options if they wish to collect anything more than the simplest one word response:

1. Require that respondents structure their feedback in some way (e.g. comma delimited) or;
2. Solicit answers one SMS at a time, with responses still needing to be confined to a reasonably limited vocabulary.

But what if you could ask a question (or several questions) in an advert, SMS or display and allow respondents to answer in free text – but still be able to analyse the data using quantitative techniques? That’s precisely what SMS Insight offers: the ability to derive structured results from free text responses. Here’s how:

1. Pose the question

Your message can be communicated via SMS, on TV or radio, in print ads or outdoor displays, on product packaging etc.

e.g. An advert asks: “What would you change about our company? SMS your name with your ideas to 35xxx”

2. Solicit feedback in plain text

Respondents simply reply via SMS – in their own words – with no fixed formats, keywords or codes required.

e.g. “I wish your staff would be more friendly. Niki”

3. Use human intelligence to crunch the responses

Our panel of mobile workers use their innate human ability to structure, classify and quantify each response by answering simple questions based on your exact analysis requirements.

e.g. “Is the response about our products, prices or service?”; “What is the respondent’s name and if possible, their gender?”

4. Integrate and act on the results

What was once raw, unusable SMS data is transformed into rich, structured information that can be used to inform business strategy, evaluate performance and identify opportunities.

e.g. “14% Products, 12% Price, 74% Service”

Interested in conducting research using SMS Insight? Please get in touch with us to discuss your requirements.

If you missed the interview which took place on Saturday 20 March 2010, you can download the podcast.

Although we had 25 participants in the pilot, 2 members of the team were not present on the Friday afternoon. The following statistics are therefore based on the remaining 23 team members who completed the self-administered survey using the Nokia 3120 mobile phones we provided for the pilot.

Age, Gender and Language

The 25 pilot participants were all from the local community of Kwanyuswa. The average age of the team members was 24 and there was an even gender split. Each of the participants had completed grade 12 and could speak fairly good English. Their first language is isiZulu but each of them studied English as a second language.

Employment history

17 of the participants (70%) had never had a full time job at the time of running the pilot. A few participants had part time jobs but were able to make the 5 hour sessions each morning.

Household information

The 17 participants that were willing to answer questions about their households have on average 7 people living permanently at home. 16 homes had stoves (94%), 14 had running water (82%), 14 had a television (82%), only 10 owned a fridge (60%) and none of the households owned a motor vehicle.

Mobile phone usage

19 of the 23 participants (82%) owned their own mobile phone (53% Nokia, 21% Samsung, 16% LG). Most participants (60%) had used MXIT (a mobile instant messaging client) in the month preceding the pilot. 9 team members (40%) had used their phones within the last month to browse the web and download pictures, music or games. The average airtime expenditure per person over the preceding 3 months was R100 per month.

Demand for mobile tasks

If employed full time in another position, the participants expressed on average that they would probably like to do Mobenzi tasks for about 3.5 hours per weekday to subsidise other income. If working only part time in another position, the desired commitment increased to 5.5 hours. Over weekends the average expected commitment was 10 hours (Including Saturday and Sunday). This works out at between 27 and 37 hours per week. 5.5 hours of concentrated work is probably the ceiling for how much time someone could spend doing Mobenzi tasks in a single day.

Everyone agreed that most Mobenzi tasks would be completed at their homes, but most participants also mentioned they would probably complete tasks while on public transport (buses and taxis) and while walking around the local community.

Thoughts on Mobenzi

The major reason the participants noted for what they liked about Mobenzi was that the work was interesting and entertaining. Only one person answered that the work was boring. The biggest challenge the team raised was that some classification tasks were ambiguous and deciding on the most appropriate answer was sometimes very difficult.

Fatigue was a problem for some participants who mentioned that their hands started hurting by the end of the day or they battled to concentrate for so long (We ran the pilot for about 5 hours each day with short breaks every hour and a longer break for lunch).

The participants were generally very excited about Mobenzi. Some of their comments are included in a related article: Feedback from pilot participants about mobile tasks

About Mobenzi

Mobenzi is a software service that empowers people to be rewarded for completing simple tasks on their mobile phones. These tasks involve certain types of problems that are difficult for a computer to solve without assistance from a real person – even someone without expert knowledge of the problem.

Find out more about how Mobenzi works

Purpose of the pilot project

For two weeks we equipped pilot participants with the Mobenzi software application installed on standard mobile phones to assess whether they could effectively complete simple business tasks using only their phones.

These were some of the guiding questions we were attempting to answer during the pilot.

• Is the concept easy to understand?
• Is the technology easy to use?
• What types of tasks are feasible?
• What types of people are most suitable for doing Mobenzi tasks?
• What is the best way to present a given task to an agent?
• How long does it take to complete different types of tasks?
• What quality should be expected in the results of completed tasks?
• What issues are involved that may affect attrition rates (fatigue, boredom etc)?
• Could the service grow through viral expansion (Can participants teach each other)?
• Based on other findings, what are the financial implications with regard to agent remuneration and the cost of the service to organisations?

Project location and venue

We ran the pilot project from the Light Providers community centre in KwaNyuswa. The area lies on the outskirts of urban development, west of the Inanda Dam, about 40 minutes outside of Durban in KwaZulu Natal, South Africa. It is one of the largest of the various tribal authorities that make up the Valley of a Thousand Hills.

Due to the gross unemployment rates in the region, and our close proximity to the area (Only 14km from our office), we selected KwaNyuswa as the location for our pilot project.

Format of the pilot

We started the first week of the pilot with 5 participants who would later act as mentors when 20 new recruits joined them for the second week. We spent the first week testing out various types of human intelligence tasks and discussing issues surrounding understanding the use of the mobile application as well as the various types of tasks themselves.

During the second week we had more participants to help work through large sets of tasks. We assigned participants various types of tasks and recorded completion times and responses for all participants so that we could crunch the data to assess what factors affect quality and efficiency.

We focused on text-based human intelligence tasks

We decided to focus on “Text to Form” tasks for the pilot project. These types of tasks involve extracting structured data from free-text.

Some examples of this type of task include:

• Categorising SMS survey responses into reportable data.
• Sentiment Rating of “Tweets” (Messages on Twitter).
• Classifying text based job and product advertisements.

For all of these tasks, we displayed a short instruction for the task, followed by the content (such as an SMS or a tweet) and then a series of questions about the content (Such as whether the SMS included a person’s name). The participant worked through each task one step at a time.

Find out about other types of human intelligence tasks

Results of the first phase of our pilot project

One of the critical factors affecting the feasibility of Mobenzi is whether or not the mobile application is easy to use for people who have had little exposure to the internet and other software applications. A quote from the summary of the first day of the pilot shows how easily the participants understood both the concept of doing work on their phones as well as how to use the application itself:

Without any instruction, most of the participants had the application open and simply started completing tasks. Although I had high expectations, I still thought there would be many questions and a fairly slow start. But within half an hour of me arriving at the venue, the participants had their heads down and were completing tasks. A few questions popped up during the day, but none that the other participants couldn’t answer themselves.

Using the software to complete tasks came very naturally and required almost zero training. From the participant comments, it is also clear that there would be a huge demand for Mobenzi tasks. I believe we could easily find thousands of Mobenzi agents who already own compatible phones within just half an hour’s drive of our offices in Hillcrest, let alone the rest of South Africa and the world.

We have not yet done much analysis on the quality or efficiency of the completed tasks, but initial assessments are very positive. Over the next few weeks we will be crunching the data to help answer some more of the questions we outlined at the start of the pilot.

The results so far have exceeded our expectations and at this stage I would guess that our biggest challenge in moving forward will be to generate a sufficient supply of tasks to keep Mobenzi agents busy.

Scaling up the pilot in April 2010

This pilot was a short 2 week project to get an early feel for what to expect. In April next year we will scale our efforts up and take on a much larger group of participants to pilot the concept further. Until then we will be tweaking the software and preparing the systems to handle the logistics of a much larger project.

We are very open to suggestions if you have any ideas for types of tasks or even real world data that we could get Mobenzi agents to process during our pilot later this year.

Firstly, for those unfamiliar with what a CAPTCHA is and the purpose of it, here’s a definition from the official CAPTCHA site:

A CAPTCHA is a program that protects websites against bots by generating and grading tests that humans can pass but current computer programs cannot. For example, humans can read distorted text as the one shown below, but current computer programs can’t:

By using real people to solve CAPTCHAs – at a very low price – spammers are able to overcome the useful function of the CAPTCHA in an economically viable fashion.

Mobenzi could be used to solve CAPTCHAs as the Mobenzi task would simply need to prompt the agent to review the image (the CAPTCHA) and respond with the actual text (thereby solving the CAPTCHA).

The example of CAPTHA solving opens the broader question: What level of monitoring and/or censorship (if any) should be imposed on those submitting tasks for processing by the Mobenzi platform?

Naturally, when Mobenzi is commercially released we will stipulate terms of use which will outline the sorts of activities which are prohibited. But is this enough? This particular example introduces some interesting issues worthy of consideration. Spamming is obviously prohibited by legislation in most countries and solving a large number of CAPTCHAs using a human labour force is rather likely to be used for this purpose, but is the process of using real people to solve CAPTCHAs automatically illegal?

Following this line of thinking, what other potentially malicious activities could be supported by the Mobenzi service and how should we guard against this? It may be infeasible, and possibly unethical to screen tasks submitted by commercial users.

One possible solution may be to use the CAPTCHA solving ability of the Mobenzi platform to prevent spammers from doing so. A small number of CAPTCHAs from major sites typically attacked by spammers could be processed by us and compared with the images requested to be processed by Mobenzi users. If a statistical match is discovered, the offending user could be sanctioned immediately.

It’s early days yet – but these types of concerns will be important should the Mobenzi service prove successful to ensure that it is used for legitimate purposes only.

Workin wit mobenzi ws great n hp w’l start soon. Al d best:-)

This is a great example of Textese (’SMS language’ involving abbreviations and slang). This comment translates to regular English as ‘Working with Mobenzi was great and I hope We’ll start soon. All the best. (said with a smile)’
[61 vs 93 characters = 34% compression].

Mobenzi is a good program/organisation which will bring many job opportunities to people, its interesting and entertaining and at the same time its challenging you to think before answering each question. Last but not least it will improve English language for many people who work with mobenzi because most of time it all about English

This was a very positive comment from one of the participants. Internally, we had discussed the potential impact Mobenzi work could have on education (such as English comprehension), but we certainly never expected participants to pick that up as a benefit during a short pilot project (it’s becoming very clear that we should stop underestimating participants).

Establish marketing strategies for mobenzi to ensure availability of tasks and more employment.

This participant seemed eager to see us succeed and offered some business advice.

I would like to work for Mobenzi.!!!

No comments, it will be a previlage working at mobenzi.

It was fun ,challenging and informative about the world that we live in.

No comment everything is new and perfect I enjoy mobenzi.

Mobenzi is very interesting and it challenges my knowledge in English and makes you think. But mostly it’s going to give us some sort of employment. THUMBS UP MOBENZI!!

This final comment sums up the sentiment of the team. I don’t think we could have expected a more positive reception to the project from the participants themselves.

1. Was the message positive, negative or neutral in reference to the brand?
2. If it was negative, was it due to customer service, taste, health or some other reason?

The work was entertaining for the participants, they completed tasks efficiently and the results seem to be very accurate.

About sentiment analysis

With the growing use of online services like Twitter, blogs and forums, there is a vast amount of publicly available information generated by everyday people about millions of different topics (companies, products, movies etc.). Knowing the sentiment of messages (e.g. whether they are positive or negative) can be extremely valuable to the people or organisations involved, especially when monitoring trends over time.

Sentiment analysis or opinion mining refers to a broad area of natural language processing, computational linguistics and text mining. Generally speaking, it aims to determine the attitude of a speaker or a writer with respect to some topic.

The rise of social media such as blogs and social networks has fuelled interest in sentiment analysis. With the proliferation of reviews, ratings, recommendations and other forms of online expression, online opinion has turned into a kind of virtual currency for businesses looking to market their products, identify new opportunities and manage their reputations.

Find out more about sentiment analysis on Wikipedia

This kind of work is well suited to Mobenzi agents

Sentiment analysis seemed like a very appropriate type of task for processing by Mobenzi agents on their phones as tweets are very short (only 140 characters). We also felt that there would be a demand for an efficient human sentiment rating service since computer algorithms face many difficulties in trying to understand the tone of human messages.

Twitter includes a lot of slang, humour, Textese and other informal language that makes automated analysis especially difficult. In a multi-cultural country like South Africa, many tweets also combine words from a variety of local languages which would make analysis very challenging to a computer.

Example Tweets that reference take-out brands

These were some of the messages included in our sample set of data from Twitter.

"Damn you debonairs" in this context is meant in jest and is not negative.

The sentiment is not obvious in this message.

The tone in this tweet changes totally at the end of the message.

The results were ‘positive’

The focus of this study was to assess issues relating to the completion of tasks. We only looked at a small sample of tweets, and could have been a lot more scientific in our approach, so the sentiment results themselves should not be taken too seriously.

There were six participants (including myself) and we each stepped through the analysis of Twitter messages that mentioned KFC, Nandos, Debonairs or McDonalds. Each task took only a few seconds to complete and the team found the work interesting and engaging. None of the participants (except myself) use Twitter themselves, but they were all very familiar with the concept and frequently use Mxit which is similar in some aspects.

One of the measurements we look at to gauge the accuracy of results, is the agreement between different participants for the same task. All six participants rated the sentiment of each tweet, so we were able to look at where our answers differed. It was very encouraging to see that most answers had 100% agreement (Especially if we exclude where participants stated that they were unsure of the sentiment). There are only a few cases where we disagreed on whether a particular tweet was positive or negative. In these cases, the majority was correct and in some cases the disagreement actually helped to balance the rating where the sentiment was ambiguous.

The summary across all brands came out at 48% positive, 35% negative and 17% neutral or unclear. Of the negative tweets, 29% were service related, 16% to do with taste, 9% health related and the rest for other reasons.

These charts show the summary of the sentiment analysis for all four brands.

In the following results, we excluded tweets that were either neutral or unclear with regard to sentiment. Out of the four brands, Nandos was clearly the favourite with 80% of tweets being rated as positive.

Breakdown of Positive versus Negative sentiment for each brand

To have a look at what people are saying right now about these brands, simply go to www.twitter.com and search for #Nandos, Debonairs, #KFC or #Mcdonalds.

Interestingly, a quick analysis of these keywords on Tweetsentiments.com (A service that attempts to automate the analysis of tweets) returns fairly similar results in terms of rank, but with some significant variations in the actual sentiment rating. Nandos: 68% positive, Debonairs 59% positive, Mcdonalds 56% positive and KFC 52% positive. The ranking of the brands is the same as our result, except that Mcdonalds moved in front of KFC with the automated analysis. This may have to do with the fact that we only looked at tweets in English and other South African dialects. Perhaps English speaking people are the least positive about Mcdonalds? Looking at some of the tweets in their data sets, I would trust our result over the automated one. Try the service out yourself at  http://tweetsentiments.com/analyze

It is no coincidence that we ended up having Nandos and Debonairs for lunch.

As with all Mobenzi tasks, we split up a large, complex challenge into small, discrete tasks which can be performed in a few minutes.

Obviously, there are limitless reasons people require translation services. One of our own immediate applications for translation is for surveys conducted using our Mobile Researcher platform. We frequently have clients embarking on multilingual studies who need questions to be available in several local languages (in South Africa alone we have 11 official languages); and need qualitative responses given in other languages available in English for analysis. Normally, surveys are designed, reviewed, debated and eventually finalised in English before they are sent for translation and back-translation. Our objective is to crowdsource this activity instead – making near real-time translations available to survey designers in Mobile Researcher at the click of a button.

In our first trial we took a client’s survey questions which we already had in English and professionally translated Zulu and converted them into two sets of translation tasks: from English to Zulu and vice versa.

As most people don’t speak Zulu (and mine isn’t exactly fluent either), I’ll use a Zulu-to-English example taken from the actual exercise:

Original English (not visible to Mobenzi agent): What is the household’s average monthly income?
Professionally Translated Zulu (shown to Mobenzi agent): Ungayilinganisela kumalini isiyonke imali engenayo ngenyanga?
Mobenzi Agent Translation (from Zulu to English): How much is the monthly income for the household?

Now, clearly the ideal would be for the original phrase to match the version coming back from the Mobenzi agent (and this did happen frequently) but in most cases, it does prove that, for the most part, crowdsourced translation works to the extent that the concept is sufficiently conveyed.

Some of the key challenges and limitations of crowdsourced translation tasks which our agents helped us identify were:

• Lack of context. Just as computer translation algorithms struggle without context – so do people. We noticed that, on several occasions, what agents had responded with made sense if read in isolation, but not within the context of the broader scope. It has to be said that this problem is exacerbated by our approach of segmenting text for translation.
• Language formality. The mobile communication culture is almost exclusively an informal one. Some agents seemed to struggle to snap out of this mode and became frustrated having to use formal language.
• Input mode. Although an obvious observation, translation requires a lot more typing than other types of tasks. Somewhat foolishly, we didn’t enable predictive text on during the first day of our translation trial which didn’t help matters.
• Multiple alternatives. Just as there are a number of ways to phrase a question in English, translation is not a one-to-one mapping. This is not a problem inherent to mobile crowdsourced translation however – two professional translators may also come up with different ways of phrasing something based on experience and personal preference.

An interesting observation is that the value in using crowdsourced translation is not only that one might not necessarily need to use traditional translation services but more that crowdsourced translation can give insight into how “normal” people understand a question in their mother tongue.

Another example:

Original English (not visible to Mobenzi agent): What is the household’s average monthly income?
Professionally Translated Zulu (shown to Mobenzi agent): Ungayilinganisela kumalini isiyonke imali engenayo ngenyanga?
Mobenzi Agent Translation (from Zulu to English): Estimate how much money do you earn a month?

In this case, the agent has incorrectly translated the question back into English by missing that the question wants to know the cumulative income of the household – not just that of the respondent. Now, this might be a mistake by the agent – but a mistake is not, in itself valueless. In this case, it can help a survey designer identify where confusion or areas of potential misunderstanding might occur in the real study. The response from the agent can guide the designer, in collaboration with their translator (professional or otherwise) in refining the question to ensure optimal wording which will make sense to the actual respondents.

We were rather pleased by the quality of translation provided by agents without any kind of moderation, statistical or other methods being applied but clearly there is still a lot of room for improvement in this area. We’ll continue working on some ideas before our main pilot takes place next year.