Te Pūnaha Matatini Whānau
For 10 weeks over the 2018-19 summer, I was involved in a project with Te Hiku Media and Dragonfly Data Science to aid in the development of a Māori voice assistant. The motivation for this project was to make Te Reo Māori more accessible and fun in the digital age.
During my internship I achieved the creation of a “box” called Rapere (translation of “Raspberry” into Te Reo Māori) containing a Raspberry Pi computer which is connected to the internet, some lights, a speaker and a microphone. This box has been coded to be continuously listening for spoken voice, and when this is detected it records what is being said until there is a longer break in the speaking (this file is overwritten each time a recording is made).
The recording is transcribed using Te Hiku Media’s Application Program Interface on koreromaori.io. The transcription that is returned to the box is compared to some key words which mean the speaker is likely asking to hear the news or to listen to the radio, or to stop playing. If these are heard, then the news or radio stream is played or stopped, and otherwise it goes back to listening for these phrases. The phrase “kia ora” lights up an LED for a few seconds. The box is able to listen for commands while playing audio, which allows the user to stop audio playing. The project was documented and all the code uploaded online to allow other developers at Te Hiku Media to progress it further and demonstrate the abilities of the Rapere box.
I experienced a great feeling of accomplishment from my work with Te Hiku Media and Dragonfly Data Science. Going from a bunch of components and an empty raspberry pi computer to having a working program with two different APIs and which plays the news on my correctly saying the appropriate phrase in Te Reo was more than I thought I would be able to achieve and I am proud of what I achieved with the help of my supervisors. I am grateful to Te Pūnaha Matatini for providing me with the opportunity to have this internship.
Cherie Vasta is a recent Statistics & Psychology graduate of Victoria University of Wellington who is starting a new role at an actuarial consulting company in Auckland. Emma enjoys applying data science techniques to all kinds of problems, especially those involving people and the way they think.
Romalee Amolic talks about her 2018-19 Te Pūnaha Matatini Summer Internship with AgResearch where she worked on a project to enhance social network analyses of biosecurity information in the New Zealand tourism industry, so that such analyses can be conducted faster and more effectively in the future.
Social Network Analysis (SNA) is a powerful data analysis technique which often helps in identifying hidden relationships and other critical information in a communication network. Data for an SNA can be collected from various sources which may result in extensive pre-processing and cleaning time as compared to the time needed for actual network analysis. Hence, this project aimed to use data carpentry to streamline the use of social research data (e.g. collected through surveys) to be able to conduct social network analyses quicker and more effectively in the future.
Better Border Biosecurity case study
The Better Border Biosecurity project, a multi-partner, cooperative science collaboration which analyses the exchange of biosecurity information in the New Zealand tourism industry, is used as a case study to develop and test methods which streamline the SNA process. The data for this project includes 154 responses from tourism providers across New Zealand who named up to 3 sources from whom they seek or receive biosecurity information. Information about the location and role of the respondents was collected. Some additional questions were also included in the survey such as the form and frequency of communication, the usefulness of information and the trust between parties.
This information was then used to perform Social Network Analysis in Gephi – a powerful interactive social network analysis tool. However, the survey data had to be first converted into a format fit for network analysis. The conventional approach for cleaning the data is discussed below.
The survey responses were collected in Excel sheets. The data pre-processing and cleaning was done manually using Excel.
Problems with the conventional approach
- It involved dealing with the data manually which was an extremely time-consuming process and needed about 1-3 weeks depending on the complexity of the data.
- The process was prone to human errors which reduced the potential of the data.
- It required skilled labour for an extended amount of time and hence, increased the costs involved.
- It led to data inconsistencies.
Hence, taking all these problems into consideration, a generic automated process was developed to clean the data as discussed in the following sections.
Data Cleaning and Pre-processing
In a network, each node represents a unique identity. Hence, the most important task in cleaning the data was to recognise and remove inconsistencies in the names which occurred due to the textual nature of responses. The following techniques were used to clean the data in Python:
- Initially, all the names were made lowercase for the analysis.
- Special characters were removed.
- Rows containing missing or no information were removed.
- Trailing or unnecessary white spaces were removed.
- Incorrect spellings were identified and removed using a spellchecker. The challenge here was to differentiate between the proper nouns (such as ‘EcoZip’) and dictionary words (such as ‘Adventures’). For example, an entity name ‘EcoZip Adventures’ was misspelled as ‘EcoZip Adventres’. A conventional spell checker would consider ‘EcoZip’ as a spelling error along with ‘Adventres’ as both the words are not found in the dictionary. Hence, a solution was developed to distinguish the proper nouns from actual dictionary words in entity names and correct spelling errors in the data.
- A custom algorithm was developed to identify abbreviations in the text and replace it with the full name. e.g. ‘DOC’ was identified as ‘Department of Conservation’.
- Several names which were written similarly but were however, the same entity, were identified and merged. This is the most significant part of the process or the most “satisfying” part, as a user described it. An example is shown below:
- All the names in the network were also compared pairwise to further remove any inconsistencies and generate a list of consistent and unique names involved in the biosecurity information exchange.
This cleaning process reduced the entities from 319 (including inconsistencies) to 139 consistent and unique entities (nodes) in the network with 335 relationships (edges) between them which were then used to generate visualisations.
Social Network Analysis
Directed maps were generated using Gephi which were then further analysed. An anonymised example of one of the social network maps generated is shown below:
The information obtained through this network analysis can now be used by biosecurity providers to better target information exchange within the New Zealand tourism industry.
- This application significantly reduced the time (from 1-3 weeks to a few hours or minutes) in cleaning and pre-processing the data before analysis.
- As a result, the costs involved in the conventional extensive processes, which involved a lot of manual effort, were also reduced.
- The new streamlined process almost eliminated the human errors involved in the manual inspection of data.
Hence, through this case study, an application was developed, which streamlines and automates all the steps starting from loading and cleaning the data up to the generation of data sheets to be used in the SNA. Although, this is currently a Python application, the development of a GUI based interactive SNA application design is currently under consideration.
I would like to acknowledge Helen Percy, my industry supervisor and Penny Payne, the social scientist at AgResearch for their invaluable support during this project.
Romalee Amolic is a Master of Applied Data Science student at the University of Canterbury. In February 2018, she completed her summer internship with AgResearch, Hamilton. She thoroughly enjoyed her internship project which involved streamlining and increasing the efficiency of the data cleaning and network map generation processes at AgResearch. She is passionate about harnessing the power of data analytics to improve the lives of people. She eagerly looks forward to applying the skills learnt, in fulfilling her aspiration of becoming a data scientist.
Congratulations to Demival Vasques Filho (Demi), our latest student to successfully defend his PhD thesis.
Demi undertook his PhD on ‘Structure and dynamics of social bipartite and projected networks’ at the University of Auckland, under the supervision of Te Pūnaha Matatini Principal Investigator Dion O’Neale. He now leaves us to take up a new position at the Leibniz Institute of European History in Mainz, Germany.
Thank you Demi for being such an active part of TPM over the years. We will miss you! Kia kaha!
— Dion O’Neale (@droneale) February 18, 2019
Congratulations to Te Pūnaha Matatini PhD student Kyle Higham, our much admired and highly active TPM Whānau past-chair and member, who successfully defended his PhD thesis recently.
Kyle undertook his PhD at the Victoria University of Wellington, researching knowledge diffusion and the dynamics of citation networks under the supervision of TPM investigators Adam Jaffe, Michele Governale and Uli Zuelicke.
Well that’s that out of the way! pic.twitter.com/2V97xYPY20
— Kyle Higham (@SpeckOnADot) November 7, 2018
Always a popular figure at TPM gatherings, Kyle now leaves us to take up an exciting role at the prestigious Ecole polytechnique fédérale de Lausanne (EPFL) in Switzerland.
Well done for all that you’ve achieved Kyle, and thank you for all your work with TPM Whānau. We hope to see you back in the future. Ka kite anō, kia kaha!
Congratulations to Samin Aref, a highly valued member of the TPM Whānau, for handing in his PhD thesis recently. Samin commenced his PhD under the supervision of TPM Investigator Mark Wilson, working on computationally intensive problems in complex networks. Samin has the honour of being the first ever TPM intern to graduate with a PhD.
Another reason to celebrate is that Samin has secured a post-doctorate position at the prestigious Max Plank Institute for Demographic Research in Rostock, Germany.
Samin flew out to Europe earlier this week. However, not before we were able to give him a proper send-off – during the TPM Whānau Retreat in Ōtaki and subsequently at TPM HQ in Auckland. Samin’s supervisor Mark Wilson and our director Shaun Hendy gave speeches recognising his contributions during his time with TPM.
Kia kaha, all the best in Germany Samin! Stay in touch and we hope to see you back in Aotearoa in the future!
Before I started working as a research assistant on the Hidden Networks project, the only woman from the history of New Zealand science I could name was Joan Wiffen, the “dinosaur lady” who discovered New Zealand’s first dinosaur fossils in Hawke’s Bay. She was a remarkable woman who contributed much to palaeontology here in New Zealand; she was also, incidentally, very white. I too am outwardly (that is, I pass as) very white. But as a mixed-race woman of Samoan descent, when I started this project I was very interested to learn about the contributions of non-Pākehā – chiefly, Māori and Pacific Island – women to science in Aotearoa. For the purposes of my research, I’ve taken “woman in science” to broadly mean a woman who has made a contribution to science in New Zealand, including both professional scientists with academic backgrounds and amateur scientists who have added to the pool of knowledge in their field, like Joan Wiffen.
The more I researched, the whiter the history of women in science in New Zealand came to look. Unsurprising really: according to Elizabeth McKinley, in 1998 just 1.5% of total employees at seven Crown Research Institutes in New Zealand identified as Māori women; there were none in management positions, and only two scientists. In ‘Finding Matilda’, Kate Hannah notes that “the historiography of science in New Zealand … tends to inadvertently reinforce [the] camouflage” of women. They are marginalized, but not absent: if you go looking, as I have, you’ll find a staggering number of women in New Zealand science from the 14th century to present-day. Yet from the beginnings of European presence in New Zealand, the overwhelming majority of these women were white. A feminist revisionist history of science aims not only to make science less male-centric (i.e. demonstrate, through promotion of women’s work both quantitatively and qualitatively, that science never has been just a man’s world) but also to make it less monochromatic (so to speak), which means celebrating the scientific achievements of brown women in New Zealand’s history, and showing that science never has been just a white world either.
In fact, the first women who made scientific contributions in Aotearoa were not Pākehā but Māori. I was delighted to learn of Whakaotirangi, who in the 1300s “was responsible for safeguarding the seed of the kūmara” as the Tainui Waka journeyed to Waikato. She was the wife of Hoturoa, the leader of the Tainui Waka migration from Hawaiki to Aotearoa, but also an important historical figure in her own right. In ‘Whakaotirangi: A Canoe Tradition’, Diane Gordon-Burns and Rāwiri Taonui explore how her importance has been diminished in post-European contact accounts of the Tainui migration. Tainui and Te Arawa traditions both speak of Whakaotirangi: she appears to be a noble and important ancestor in the history of both iwi. While she is most remembered for bringing kūmara to Waikato, she was also responsible for a number of other plants brought from Hawaiki. On arrival in Waikato, Whakaotirangi built gardens in which she experimented with growing and tending to a variety of plants, both for sustenance and medicinal purposes. She discovered how to make the kūmara, which had come from a much warmer climate, grow in the cooler land her people had settled. Her work was crucial for the establishment of the Tainui people: it provided them with a reliable food supply as they adjusted to life in a new land. She was also involved in commissioning, building and launching the Tainui canoe. Her profile on the Royal Society of New Zealand website, as part of their series 150 Women in 150 Words, credits her as “one of New Zealand’s first scientists”.
Around the middle of the 1400s, another important ancestor of the Waikato people appeared. Kahu (also known as Kahupeka, Kahupekapeka, Kahukeke, or Kahurere) was a Tainui woman who experimented with plants – such as harakeke, koromiko, kawakawa and rangiora – as medicinal remedies. She did so during her great journey: walking inland through the King Country while grieving the death of her husband (who in some accounts is Rakataura; in others Uenga). She gave names to different sites along her journey (such as Te Manga-Wāero-o-Te Aroaro-ō-Kahu – ‘the stream in which Kahu’s dogskin cloak was washed’) – these names tell the story of her journey and preserve the history of the land. At some point during her journey she was ill, which may have been why she sought out plants for their medicinal properties. Unfortunately there are many different versions of Kahupeka’s story, and in them there are few mentions of her medicinal experimentations with indigenous flora. In some versions Rakataura doesn’t die, and he and Kahu traverse the countryside naming places together, as explorers.
In Māori culture, practitioners or experts in any skill or art are known as tohunga. The Tohunga Suppression Act 1907 made tohunga status a punishable offence. The Act was repealed only in 1962, and so much of the knowledge surrounding this customary way of knowing has been suppressed – my search for tohunga wahine (female practitioners) who might count as women of science has not produced significant results. However, it is worth noting that the sources I accessed relied upon the written record. Other sources, such as Māori oral histories, may be much more fruitful.
The next Māori woman in science that I was able to find wasn’t born until the 19th century. Makereti Papakura (Margaret Pattison Thom; she also went by Maggie and was of Te Arawa and Tuhourangi iwi) was born to a Māori mother and an English father in the Bay of Plenty in 1873. She was raised by her mother’s aunt and uncle in Parekarangi, a rural area. She didn’t learn English until she was ten years old, speaking only Māori until her father took over her education. After her schooling, Papakura moved to Whakarewarewa, where she became an accomplished tourist guide. She gave herself the surname Papakura after a nearby geyser when a tourist she was guiding asked if she had a Māori surname. Clearly, the name stuck. In 1891 she married surveyor Francis Joseph Dennan; they had one child together before divorcing in 1900. In 1905 she wrote Guide to the hot lakes district. Papakura travelled to England in 1912, and married Richard Charles Staples-Browne. She had first met Staples-Brown when he was on a tour of New Zealand, and had reconnected with him while she was part of a Māori tour party in England. They divorced in 1924, but Papakura remained in England and in 1926 she enrolled at Oxford University, studying a BSc in anthropology. She died on April 16, 1930, only two weeks before her thesis, The old-time Māori – in which Papakura combined customary knowledge with scholarly conventions – was due to be examined. It was published posthumously, eight years later. Her thesis covers Māori social and familial structures, housing, weaponry and relationship with fire. She was meticulous in her writing, and wrote letters to her people in New Zealand during her drafting process, to ensure her account was as accurate as possible.
Bessie Te Wenerau Grace (1889-1944; Ngāti Tūwharetoa) was the first Māori woman university graduate, graduating from Canterbury University with a BA in 1926. She was the granddaughter of Ngāti Tūwharetoa chief Horonuku Te Heuheu. She then went on to receive an MA with first-class honours in modern languages from London University. In London she also became a nun, Sister Eudora. She worked as headmistress of St Michael’s School in Melbourne. In 1945, Dame Mira Szászy (1921-2001; Ngāti Kurī, Te Rarawa, and Te Aupōuri), a prominent Māori leader, became the first Māori woman to graduate with a degree from the University of Auckland. She went on to complete a postgraduate diploma in social sciences from the University of Hawaii and worked hard to improve the welfare of Māori women throughout her life. In 1949, Rina Winifred Moore (1923-1975; Ngati Kahungunu, Rangitane and Te Whanau-a-Apanui) graduated from the University of Otago with a Bachelor of Medicine and Bachelor of Surgery – and in so doing, became the first Māori woman doctor in New Zealand. In her career she worked to improve public perceptions of the mentally ill and was one of the first doctors in New Zealand to prescribe the contraceptive pill.
It has been harder for Māori and Pacific Islanders to enter scientific professions, as they are forced to combat social prejudices that expect them to fail – that tell them this is not where they belong. It has been harder for women to enter scientific professions because, again, they have to fight against the social biases that tell them ‘this is not your world’. Until the late 20th century, many women were expected to give up their careers when they married – motherhood and the domestic sphere became their full-time responsibilities. Some women chose to remain unmarried and childless in pursuit of scientific careers, while others stopped working when they married. Māori and Pacific women have to fight both gender and racial biases for their place in the world of science. This has been the case throughout the post-contact history of Aotearoa, and continues to be so.
Today, there are increasing numbers of Māori and Pacific Island women in science, with some of them working at the intersection of traditional knowledge and western science. Dr Ocean Mercier (Ngāti Porou) is a Senior Lecturer in Māori Science (the intersection of western science and mātauranga Māori) at Victoria University of Wellington. She has a PhD in Physics and was awarded the New Zealand Association of Scientists (NZAS) inaugural Lucy Cranwell Medal (previously the Science Communicators’ Medal) in 2017. Science researcher Hokimate Harwood (Ngāpuhi) combines western scientific and Māori customary knowledge in her research of the feathers in kahu huruhuru (feather cloaks). Her use of microscopy to identify the origins of feathers used in precious cloaks has been pioneering. She is a Bicultural Science Researcher at Te Papa. Her sister, Dr Matire Harwood (Ngāpuhi; PhD MBChB), is a Senior Lecturer at the University of Auckland Medical School and has done crucial research into indigenous healthcare throughout her career. Her efforts have been widely recognised, and in 2017 she was awarded a fellowship to the L’Oréal UNESCO For Women in Science programme.
Victoria University science educator Dr Hiria McRae (Te Arawa, Tūhoe, Ngāti Kahungunu) has created and developed a new educational model aimed at raising Māori students’ engagement in high schools. Through her research projects she has made important contributions to the field of Māori education.
Victoria University astrophysicist, science lecturer and research fellow Dr Pauline Harris (Rongomaiwahine and Ngāti Kahungunu), who has a PhD in astroparticle physics, is a key figure in the revitalisation and teaching of Māori astronomy. She is also involved in the search for extra-solar planets. Connected to Harris’s Māori astronomy programme is Pounamu Tipiwai Chambers, an undergraduate student at Victoria University who has employed Māori astronomical and navigational knowledge in undertaking waka voyages across the Pacific.
Another remarkable young woman, Alexia Hilbertidou (of Greek and Samoan descent), has founded GirlBoss New Zealand, an organisation aimed at the empowerment of young women in STEM studies after she felt alienated as the only girl in her year thirteen physics for engineering class. She was also part of NASA’s SOFIA project, making her the youngest person ever to be part of a NASA mission.
My blog post aims to contribute towards the unmasking of Māori and Pacific women’s contributions to science in both historical and contemporary landscapes. We are already seeing some important changes: many Māori women in science today combine customary and scientific knowledge to great success, a road paved by Makereti Papakura and her BSc thesis. However, Māori and Pacific women are still dramatically under-represented in fields of science, particularly at senior and management levels. It is therefore important that we keep up the momentum of positive change not only by looking forward but also by looking back: the successes of past figures provide an encouraging bevy of ‘shoulders to stand on’ for women in science today.
This post was written as part of my summer scholarship research on the Hidden Networks project, supervised by Rebecca Priestley and Kate Hannah.
If you’re interested in learning more about the women I’ve mentioned, you might enjoy some of these sources:
- Colenso, William, ‘Contributions towards a better Knowledge of the Maori Race’, in Transactions and Proceedings of the Royal Society of New Zealand 1868-1961, Vol. 14, 1881, pp. 33-48. http://rsnz.natlib.govt.nz/volume/rsnz_14/rsnz_14_00_000690.html
- ‘Dr Ocean Mercier wins prestigious Science Communicator’s Medal.’ https://www.victoria.ac.nz/news/2017/11/dr-ocean-mercier-wins-prestigious-science-communicators-medal
- GirlBoss New Zealand. https://www.girlboss.nz/
- Hilbertidou, Alexia, ‘NASA SOFIA Experience’, U.S. Embasssy & Consulate in New Zealand. https://nz.usembassy.gov/alexia-hilbertidou-nasa-sofia-experience/
- ‘Hokimate Harwood – Identifying feathers’, Museum of New Zealand Te Papa Tongarewa. https://collections.tepapa.govt.nz/topic/3657
- Mack, Ben, ‘How Dr Matire Harwood is addressing inequities in healthcare for indigenous people’, Idealog, 3 November 2017. https://idealog.co.nz/etc/2017/11/how-dr-matire-harwood-addressing-inequities-healthcare-indigenous-people
- ‘Māori science education model developed’, Radio New Zealand, 28 August 2015. https://www.radionz.co.nz/news/te-manu-korihi/282635/maori-science-education-model-developed
- McKinley, Elizabeth. ‘Brown Bodies, White Coats: Postcolonialism, Māori women and science’, in Discourse: Studies in the Cultural Politics of Education 26 no. 4, 2005, pp. 481-496. http://www-tandfonline-com.helicon.vuw.ac.nz/doi/full/10.1080/01596300500319761?scroll=top&needAccess=true
- Morton, Jamie, ‘Royal Society tackling diversity issues’, New Zealand Herald, 26 October 2016. http://www.nzherald.co.nz/nz/news/article.cfm?c_id=1&objectid=11736157
- Morton, Jamie, ‘Q&A: NZ science’s own ‘Hidden Figures’, New Zealand Herald, 24 January 2017. http://www.nzherald.co.nz/nz/news/article.cfm?c_id=1&objectid=11787672
- Northcroft-Grant, June. ‘Papakura, Makereti’, Dictionary of New Zealand Biography, first published in 1996. https://teara.govt.nz/en/biographies/3p5/papakura-makereti
- ‘Researcher to teach traditional Māori astronomy’, Radio New Zealand, 17 June 2013. https://www.radionz.co.nz/news/te-manu-korihi/137868/researcher-to-teach-traditional-maori-astronomy
- Royal, Te Ahukaramū Charles, ‘Waikato tribes – Ancestors’, Te Ara – the Encyclopedia of New Zealand, http://www.TeAra.govt.nz/en/waikato-tribes/page-3
- Shaw, Aimee, ‘Meet Alexia Hilbertidou, the 18-year-old founder of GirlBoss and the youngest person to be involved with Nasa’s Sofia mission’, New Zealand Herald, 11 July 2017. http://www.nzherald.co.nz/business/news/article.cfm?c_id=3&objectid=11888757
- Tipiwai Chambers, Pounamu. ‘Te Ara Tauira’ in Salient, 1 May 2017. http://salient.org.nz/2017/05/te-ara-tauira-4/
Beth Rust is a BA(Hons) history graduate from Victoria University of Wellington. For her Honours thesis she researched the writings of Christine de Pizan, a 15th-century humanist and early defender of womankind. This past three months she has been working as a research assistant on the project ‘Hidden Networks: hybrid approaches for the history of science’. Beth is just about to start a job in the public service, and she is very excited to take the skills she has learned from her summer research into her new role. She loved being a summer scholar.
I wrote this algorithm to classify people by gender, but one of the biggest things I learned was how machine learning can reinforce racism and perform poorly on ethnic minorities.
Machine learning – or programs that are able to learn from and improve on past experience and data – is often accused of reinforcing human biases such as racism and sexism. However, it can be a bit unclear how exactly this happens.
How does an automatic soap dispenser fail to recognize black people’s hands? How does image recognition software come to classify people in kitchens as women, regardless of their actual gender? How does artificial intelligence that seeks to predict criminal recidivism produce results that are consistently biased against black people?
This walk-through hopes to give you a bit of an insight into one example of racism in machine learning, and how this comes to be.
The algorithm will be used as part of research into gender equity in STEM fields in New Zealand. A lot of information about who works in certain research centres or who graduated from university is publicly available online (for example, here are university records from NZ between 1870 and 1961), but it doesn’t explicitly include their gender. While a person reading the information can usually guess their gender quite easily and with a high degree of accuracy, it’s obviously very impractical to read and classify thousands or hundreds of thousands of observations. This is where this algorithm hopes to simplify and speed up the process of identifying women in STEM fields.
Training and testing data: Selecting appropriate data
Getting good data for the training and test sets is a really important part of machine learning. Your model is only as good as the data you train and test it on, so getting this right is key.
The starting point of my dataset is the 100 most common names for boys and girls born in New Zealand in each year, going back to 1954. One major drawback of this dataset is that it only includes people born in New Zealand, not those that emigrated here. This means the dataset is almost exclusively made up of Anglo-Saxon names, and does not reflect New Zealand’s large Asian and Pacific populations.
🎉 New #data alert! 🎉
We’ve just updated our figures on a topic that’s always popular – #baby names 👶🏽
Charlotte and Oliver topped the 2017 charts, but we’re sure you’ll spot plenty of other familiar names. Know anyone with a name that made the top 50? pic.twitter.com/A1eHH4kGq5
— Figure.NZ (@FigureNZ) January 30, 2018
It also doesn’t include any Māori names, presumably because the Māori population isn’t large enough for these names to make the top 100 list. I’ve tried to remedy this by adding the top 20 Māori names for boys and girls from several years to the dataset. However, 91% of the training dataset is still made up of Anglo-Saxon names, while only 9% is made up of Māori names.
These biases in the training dataset mean that the model is likely to recognize the patterns that indicate gender in Anglo-Saxon names, while not picking up on patterns that indicate gender in the names of other cultures. The same biases in the testing dataset mean that the accuracy of the model probably only applies to Anglo-Saxon names, and that it may do much worse on names of other nationalities.
Selecting useful features for the algorithm
It’s important to consider what features would be most useful in predicting the desired classes. I started off by using the last letter of each name to predict gender. Most Anglo-Saxon names for men end with a consonant, while most Anglo-Saxon names for women end with a vowel.
There are also some pairs of letters that are more common for one gender than the other. For example, the last letter ‘n’ is indicative of a male name (e.g. Brian, Aidan, John), but the suffix ‘yn’ is indicative of a female name (eg. Robyn, Jasmyn). Because of this, using both the last letter of each name and the suffix as features results in higher accuracy than just using the final letter. This gave me an accuracy of about 73% on a testing dataset that includes both Anglo-Saxon and Māori names.
This overall accuracy is lower than it would have been on a testing dataset made up of only Anglo-Saxon names because these features don’t perform as well with names of other origins. In a New Zealand context, this causes the most problems with Māori names. Most Māori names end in vowels, regardless of gender (examples of male Māori names include Tane and Nikau, while female Māori names include Aroha and Kaia). This means this particular feature doesn’t do a very good job with names of Māori origin.
The same problem would likely apply to other ethnicities, too. For example, Japanese, Chinese, Vietnamese, Italian and Hispanic names all often end in vowels, regardless of gender.
Imbalanced classes and the problems they cause
Imbalanced classes, or classes that are very different in their size, can also create problems for machine learning algorithms. In this case, ethnicity is an imbalanced class that is likely to influence people’s names. In the 2013 census, 74% of New Zealanders identified as European, 15% as Māori, 12% as Asian and 7% as Pacific. (Note that Statistics New Zealand allows you to identify with more than one ethnicity, therefore these numbers don’t add up to 100%).
Imbalanced classes often result in high accuracy within the majority class (in this case, European) and low accuracy within the minority classes (Māori, Asian and Pacific). This particular algorithm has an overall accuracy of about 73%. The accuracy within Māori names is about 69%, while the accuracy within European names is 75%.
The class imbalances in the data explain why the overall accuracy may not be a very good way of assessing whether the algorithm is working well. As well as checking the accuracy within each subgroup, it can be a good idea to look at precision and recall for more information on where the algorithm is doing well and where it’s doing poorly.
Precision tells us how much of a classified group actually belongs to that group. In this case, for example, precision of female names is the percentage of names classified as female that are actually female. It is calculated by dividing the number of true positive (number of women classified as female) by all positives (number of women and men classified as female).
Recall is the percentage of a particular group that has been classified as belonging to that group. For example, recall of male names is the percentage of male names that have been classified as male. Recall is calculated by dividing the number of true positives (number of men classified as male) by the number of true positives and false negatives (number of men classified as female).
The tables below show the precision, recall and a couple of other metrics on how well the algorithm is doing. The differences between the overall table and the tables by ethnicity show that it’s likely that this algorithm is systematically worse with non Anglo-Saxon names, specifically Māori names in this instance.
For Māori names only:
Here we can see that both precision and recall is very low for male Māori names. This means that only a small percentage of the names classified as being male actually are male (low precision) and an even smaller percentage of male Māori names have been classified as being male (low recall).
This is probably because most Māori names end in vowels, regardless of their gender. The algorithm does alright on female Māori names, because it has seen many instances of female names ending in vowels before. But it hasn’t seen many male names ending in vowels, so it fails to classify most of these names correctly.
For European names only:
Because machine learning algorithms with imbalanced classes usually do worse in the smaller classes, they can further marginalise minority groups by routinely misclassifying them or failing to take into account patterns that are unique to the smaller group. In this example, this is likely to be the case with ethnic minorities.
It seems that this algorithm is likely to really only do a good job on Anglo-Saxon names. This limits the situations in which it would be appropriate to use it, and risks reinforcing Eurocentricity and a focus on whiteness.
This example shows how difficulties in getting hold of representative datasets, selecting features and unbalanced classes can cause algorithms to perform poorly on minority groups. These are only a couple of the many ways machine learning can contribute to the marginalisation of minorities, and it’s important to consider how this might happen in the particular algorithm you’re working on.
The consequences of bias in machine learning can range from the irritation of not being able to get soap out of an automatic dispenser, to the devastation of being given a longer prison sentence. As these algorithms become more and more ubiquitous, it is essential that we consider these consequences in the design and application of machine learning.
See this paper for a more detailed look at how imbalanced classes affect machine learning algorithms.
Emma Vitz is a recent Statistics & Psychology graduate of Victoria University who is starting a new role at an actuarial consulting company in Auckland. Emma enjoys applying data science techniques to all kinds of problems, especially those involving people and the way they think.
In the immortal words of Vanilla Ice – Stop, collaborate and listen. Collaboration is a cornerstone of modern science and with flight tickets cheaper than ever before and the internet effectively eliminating the expense of correspondence, academics and researchers are looking further afield and reaching more contemporaries across the globe. However, different institutions have different facilities and research focuses, not everyone speaks the same language, and so perhaps these researchers may be picky when it comes to who they work with. It raises the question of whether they do have a preference in collaborator based on affiliation and, if so, can this preference be measured and distilled into cold, hard data?
Of course they do, and of course it can be. More to the point, why?
Arguably the most tangible and conveniently quantifiable means in which academic collaboration manifests is in scientific papers and articles, typically with several authors from varying affiliations. A notable drawback in previous studies on research collaboration is that the measures used (such as the fractional count detailed in Nature Index) consider results for each institution, rather than individual academic, and disregard the size of each institution; as a result, smaller and younger institutions may stack up unfavourably compared to those that are more established and larger. For example, take a look at how the eight New Zealand universities compare against each other:
- The nodes representing each university are weighted by their respective output (total number of co-authored papers by academics affiliated with these universities).
- The links connecting universities to each other are weighted by the number of papers co-authored by researchers from both institutions.
- The higher the link weight, the more that the connected universities are attracted to each other.
The skewing effect that university size has on this network is pretty apparent from how Lincoln University has much fewer co-authorships with Victoria University and University of Waikato than with the rest of the network, given its relatively small output. Also of note is that the University of Auckland and AUT have a much lower link weight than one would expect for two universities across the street from each other, yet the University of Auckland and the University of Canterbury have a much stronger link despite being at opposite ends of the country.
First, to address the effect of institution output. We do this using something we call the revealed comparative preference (RCP) of an institution i for collaborating with institution j:
where Xij is the number of co-authorships between i and j, Xi is the total number of papers co-authored by i with other institutions in the data set, and X is the total number of co-authorships between all the institutions in the data set.
Plainly speaking, it’s a measure of whether two institutions are doing more than collaborating than we might expect with each other relative to their tendency to collaborate with the other universities in the data set. If Pij > 1 , then universities i and j share more co-authorships than we expect relative to the other institutions in the data set, so we say they have a comparative preference for collaborating with each other. Conversely, Pij < 1 indicates that the two universities are doing less than we might expect.
Anyway. Here’s the NZ university network revised with the links now weighted by their corresponding RCP values:
Better. Here it’s apparent that AUT has a stronger link with Auckland Uni in addition to Lincoln and Waikato, and it should be pointed out that University of Auckland, AUT and Massey University are also closer to each other in the network, bearing in mind that all three have campuses within Auckland.
Now with a working measure, we move on to a larger sample. Bring on the Australians.
Clearly the Tasman Sea has a solid effect on the way New Zealand based researchers connect with those based in Australia; the links within the NZ cluster of universities have greater RCP weightings than those within the Australian cluster, implying a preference for domestic rather than trans-Tasman co-operation. Another feature to consider is that the Australian universities in the same states are grouped together, which is consistent with the idea that geographical proximity plays a significant part in a researcher’s choice of collaborator.
It would only be natural to wonder how academics interact on a global scale – do we ever grow out of talking almost exclusively to our friends and shun outsiders in some weird, grown up, Mean Girls-esque collection of cliques?
From observing how the Dutch and German institutions are grouped together, we might conclude that the language barrier is a large hurdle to overcome when jointly writing scientific literature – this also seems apparent from the Chinese-Hong Kong cluster, as well as Korean and Japanese institutions as well. But languages also tend to cluster geographically, so it is hard to disentangle the effect of language from distance.
It’s no question that with the constant progress of technology, connecting with people is becoming less costly. However, there are factors remaining that impede the prospect of a totally connected scientific community, some of which have been speculated on here. Of course pictures and hand waving don’t constitute a solid argument, but a thorough analysis of these factors and their effect on university collaboration will be in store for you, dear reader.
In the meantime, perhaps one should learn German, or Mandarin, or Dutch, or even Japanese. It’s not that hard.
About the data visualisations
In order to make the larger graphs efficient enough to be used in browser, the amount of connections a node could have to other nodes was limited to its top four RCP values. This change had no significant effect on the clustering observed when the full connection matrix was used. The change was only implemented for the QS, ANZAC and benchmark data sets.
Bonnie Yu is a research assistant at Te Pūnaha Matatini and a member of Te Pūnaha Matatini’s Whānau group for emerging scientists. Her research projects focus on university collaboration networks.
The data visualisations of this post were prepared by fellow research assistant, Nickolas Morton.
What are you going to do after you finish your PhD? Where do you want to go? Are you going to become a lecturer? These are all questions that I field on a regular basis. Rather than going with my instinctive response of “What the hell? I don’t even know what my PhD is about yet!”, I usually say something like “I don’t know, but hopefully something in conservation or consulting”. Apparently this puts me in the minority of PhD students in that I do not desire to go into academia.
This was a topic discussed at the New Zealand Association of Scientists conference I attended on the 26th April; you can also read about it in my previous blog post. One of the speakers referenced the Royal Society report where it stated that while about half of PhD students continue on with research, becoming early career researchers, most end up leaving academia for work in industry. This is despite most PhD candidates desiring a job in academia at the outset of the project. The question asked at the conference is how can we, as the scientific community, support PhDs and Post Docs so that if an academic career does not pan out they can successfully and relatively painlessly transition into industry? As one of the members of the emerging researchers panel said, when she was faced with the current situation, it is not unusual to feel like the best option is just to “give up”.
I am lucky in that I have a great team of supervisors (I have 4 ± 1 supervisors) who want my PhD to be more about preparing me for future work rather than me just churning out papers. They have suggested that I take opportunities to learn skills that will be useful in industry and that I take time to build connections inside and outside of academia. However, not everyone is as lucky in having such excellent supervisors. I have heard horror stories about supervisors who refuse to meet with their students and those who take no role in preparing the student for the future. What can we do for these students without supervisor support?
This is a place where student-led organisations can step in. The Te Pūnaha Matatini Whānau committee is well aware of these trends and are currently working on a number of projects to address this. The Whānau has connected with industry partners such as data analytic companies. The intention is for TPM Whānau members to be eligible to undertake internships at the companies. This will teach the members new skills and give experience that will be valuable in industry. We are also organising a data debate on the issues of data privacy between industry members and Te Pūnaha Matatini.
Ultimately however, no matter how supportive the supervisor is, it is up to the student to make sure that they obtain the experience and skills they need. As one of my supervisors said, “if you are smart enough to get to PhD level you are smart enough to look after yourself”.
With that I will sign off and go look after myself.
Jonathan Goodman is a Te Pūnaha Matatini Whānau committee member.
By Jonathan Goodman
Never do things by halves, jump in the deep end, give it a go, eat your vegetables, trust your supervisors. This is all good advice and I now realise I must have taken it, having presented at the first conference I have ever attended, then attending another conference three days later run by an organisation I had never heard of before. I have also joined the Te Pūnaha Matatini Whānau committee based solely on my supervisor’s advice. Before I go on, I must admit that all of these actions have proved to be worthwhile and rewarding.
The first conference was the Te Pūnaha Matatini cross-theme hui. This was the first Te Pūnaha Matatini gathering I have attended since joining the Centre of Research Excellence as a PhD student at the start of the year. The hui consisted of a series of short talks, including my first at a conference, interspersed with four rounds of the “Research Knockout” – a game designed by Alex James. The game started with the creation of teams of 3-5 researchers from Te Pūnaha Matatini’s three research themes. Each team then generated a potential research project. Each round of the knockout consisted of pairing up the groups and amalgamating their ideas into an enhanced version. This continued until there were just two groups remaining. In the grand finale, there was a final presentation followed by a vote. The winning research topic was ‘Measuring the impact of the communication of science’.
The question of science outreach also came up at the conference run by the New Zealand Association of Scientists (NZAS). The conference was held at Te Papa in Wellington and celebrated the 75th anniversary of the Association. The conference had a selection of engaging speakers looking at the role of scientists in the past, the present, and into the future. A number of speakers talked about science communication.
One of the presenters, Simon Nathan, spoke about James Hector and how he effectively pushed the cause of New Zealand science, through his role of Chief Government Scientist, by constantly reminding politicians about the value of science. Rebecca Priestley talked about how science outreach was different back in the days of the Department of Scientific and Industrial Research (DSIR). Instead of scientists engaging in outreach programs, interested journalists and citizens would phone and be able to speak directly with the scientist who was in the best position to answer their queries. Te Pūnaha Matatini’s own Shaun Hendy presented on how social media is currently the only way scientists are able to directly communicate with the population without the risk of their message being obscured. His three guidelines for public engagement were very apt.
1) Not be d!@#s
2) Get on social media
3) See rule number 1.
The other major theme of the conference was the structure of the pathways inside and outside academia for emerging researchers. I will touch on this in another blog post on the Te Pūnaha Matatini Whānau page.
Having had a rewarding weekend forming connections with talented scientists, and with the science community as a whole, I will sign off hoping that I have followed Shaun’s rules.