5

LinkedIn says it reduced bias in its connection suggestion algorithm

 3 years ago
source link: https://venturebeat.com/2021/08/05/linkedin-says-it-reduced-bias-in-its-connection-suggestion-algorithm/
Go to the source link to view the article. You can view the picture content, updated content and better typesetting reading experience. If the link is broken, please click the button below to view the snapshot at that time.
neoserver,ios ssh client

LinkedIn says it reduced bias in its connection suggestion algorithm

Image Credit: LinkedIn

Data-driven creative for sensitive verticals

With big privacy changes, creative has become even more important with verticals like health and wellness and finance. Learn how to make data the backbone of your campaigns.

Register for free now

All the sessions from Transform 2021 are available on-demand now. Watch now.


In a blog post today, LinkedIn revealed that it recently completed internal audits aimed at improving People You May Know (PYMK), an AI-powered feature on the platform that suggests other members for users to connect with. LinkedIn claims the changes “level the playing field” for those who have fewer connections and spend less time building their online networks, making PYMK ostensibly useful for more people.

PYMK was the first AI-powered recommender feature at LinkedIn. Appearing on the My Network page, it provides connection suggestions based on commonalities between users and other LinkedIn members, as well as contacts users have imported from email and smartphone address books. Specifically, PYMK draws on shared connections and profile information and experiences, as well as things like employment at a company or in an industry and educational background.

From automated Pizza delivery, to better Health outcomes; the Role of Intelligent Virtual Agents in the New Digital Workforce 1

PYMK worked well enough for most users, according to LinkedIn, but it gave some members a “very large” number of connection requests, creating a feedback loop that decreased the likelihood other, less-well-connected members would be ranked highly in PYMK suggestions. Frequently active members on LinkedIn tended to have greater representation in the data used to train the algorithms powering PYMK, leading it to become increasingly biased toward optimizing for frequent users at the expense of infrequent users.

“A common problem when optimizing an AI model for connections is that it often creates a strong ‘rich getting richer’ effect, where the most active members on the platform build a great network, but less active members lose out,” Albert Cui, senior product manager of AI and machine learning at LinkedIn, told VentureBeat via email. “It’s important for us to make PYMK as equitable as possible because we have seen that members’ networks, and their strength, can have a direct impact on professional opportunities. In order to positively impact members’ professional networks, we must acknowledge and remove any barriers to equity.”

Biased algorithms

This isn’t the first time LinkedIn has discovered bias in the recommendation algorithms powering its platform’s features. Years ago, the company found that the AI it used to match job candidates with opportunities was ranking candidates partly on the basis of how likely they were to apply for a position or respond to a recruiter. The system wound up referring more men than women for open roles simply because men are often more aggressive at seeking out new opportunities. To counter this, LinkedIn built an adversarial algorithm designed to ensure that the recommendation system includes a representative distribution of users across gender before referring the matches curated by the original system.

In 2016, a report in the Seattle Times suggested LinkedIn’s search algorithm might be giving biased results, too — along gender lines. According to the publication, searches for the 100 most common male names in the U.S. triggered no prompts asking if users meant predominantly female names, but similar searches of popular female first names paired with placeholder last names brought up LinkedIn’s suggestion to change “Andrea Jones” to “Andrew Jones,” “Danielle” to “Daniel,” “Michaela” to “Michael,” and “Alexa” to “Alex,” for example. LinkedIn denied at the time that its search algorithm was biased but later rolled out an update so any user who searches for a full name if they meant to look up a different name wouldn’t be prompted with suggestions.

Recent history has shown that social media recommendation algorithms are particularly prone to bias, intentional or not. A May 2020 Wall Street Journal article brought to light an internal Facebook study that found the majority of people who join extremist groups do so because of the company’s recommendation algorithms. In April 2019, Bloomberg reported that videos made by far-right creators were among YouTube’s most-watched content. And in a recent report by Media Matters for America, the media monitoring group presents evidence that TikTok’s recommendation algorithm is pushing users toward accounts with far-right views supposedly prohibited on the platform.

Correcting for imbalance

To address the problems with PYMK, LinkedIn researchers used a post-processing technique that reranked PYMK candidates to decrement the score of recipients who’d already had many unanswered invitations. These were mostly “ubiquitously popular” members or celebrities, who often received more invites than they could respond to due to their prominence or networks. LinkedIn thought that this would decrease the number of invitations sent to candidates suggested by PYMK and therefore overall activity. However, while connection requests sent by LinkedIn members indeed decreased 1%, sessions from the people receiving invitations increased by 1% because members with fewer invitations were now receiving more and invitations were less likely to be lost in influencers’ inboxes.

As a part of its ongoing Fairness Toolkit work, LinkedIn also developed and tested methods to rerank members according to theories of equality of opportunity and equalized odds. In PYMK, qualified IMs and FMs are now given equal representation in recommendations, resulting in more invites sent (a 5.44% increase) and connections made (a 4.8% increase) to infrequent members without majorly impacting frequent members.

“One thing that interested us about this work was that some of the results were counterintuitive to what we expected. We anticipated a decrease in some engagement metrics for PYMK as a result of these changes. However, we actually saw net engagement increases after making these adjustments,” Cui continued. “Interestingly, this was similar to what we saw a few years ago when we changed our Feed ranking system to also optimize for creators, and not just for viewers. In both of these instances, we found that prioritizing metrics other than those typically associated with ‘virality’ actually led to longer-term engagement wins and a better overall experience.”

All told, LinkedIn says it reduced the number of overloaded recipients — i.e., members who received too many invitations in the past week — on the platform by 50%. The company also introduced other product changes, such as a Follow button to ensure members could still hear from popular accounts. “We’ve been encouraged by the positive results of the changes we’ve made to the PYMK algorithms so far and are looking forward to continuing to use [our internal tools] to measure fairness to groups along the lines of other attributes beyond frequency of platform visits, such as age, race, and gender,” Cui said.

VentureBeat

VentureBeat's mission is to be a digital town square for technical decision-makers to gain knowledge about transformative technology and transact.

Our site delivers essential information on data technologies and strategies to guide you as you lead your organizations. We invite you to become a member of our community, to access:

  • up-to-date information on the subjects of interest to you
  • our newsletters
  • gated thought-leader content and discounted access to our prized events, such as Transform 2021: Learn More
  • networking features, and more
Become a member

Data quality, COVID response, saving the coral reefs and more during Transform’s Data, Analytics, & Intelligent Automation Summit

VB StaffJuly 13, 2021 07:05 PM

Data-driven creative for sensitive verticals

With big privacy changes, creative has become even more important with verticals like health and wellness and finance. Learn how to make data the backbone of your campaigns.

Register for free now

Our Data, Analytics, & Intelligent Automation Summit at Transform 2021 on Tuesday took a deep dive into how data, analytics, and intelligent automation can help the greater good, the bottom line, and more.

The day, presented by Accenture, kicked off with the Big Bytes in AI & Data breakfast, presented by Accenture. Leaders from Accenture, American Express, Opendoor, Evernorth, and Google ultimately agreed that the quality of the data under their AI solutions is non-negotiable.

As Valerie Nygaard, product lead at Google Duplex, said, “You can make tons of tech innovations, but so much of the time they rely on the quality of the data, that accuracy, the normalization, the processing, and the handling.”

The American Express credit and fraud risk group uses models powered by machine learning to monitor $1.2 trillion in charges annually around the world, and return 8 billion risk decisions in real time, said Anjali Dewan, vice president of risk management, consumer marketing and enterprise personalization decision science at American Express.

“Having the discipline to make sure that the quality of that data is consistent, starting from evaluation when you put it into production, is a key competitive advantage,” she explained.

Opendoor’s valuation models, which service more than 90,000 customers, and enable more than $10 billion in real estate over 30 markets, are only as worthwhile as their data input, said co-founder and CTO Ian Wong. To ensure coverage and accuracy, they’ve built custom inspector apps that use a human expert to collect first-party data and then input it back into their central repository in real time.

It takes time to collect and manage data, ensure it is high quality and governed, and then organize it to drive insights, said Mark Clare, enterprise head of data strategy and enablement at Evernorth/Cigna. But the new agile, collaborative processes and visual-based discovery Cigna helped one financial services company implement led to the company discovering an eight-figure attrition risk in less than 30 minutes.

One big takeaway for Ahmed Chakraborty, global managing director, applied intelligence North America lead at Accenture, is that when you take a data-driven journey in the enterprise, it’s a change journey, and a big part of the change is to drive adoption.

“I call this the last-mile connection,” he said. “Literacy around data is critical. Elevating the entire acumen of your enterprise to understand data, understand what you can do with data, is so critical in the long-term journey to drive adoption and the change in your culture.”

“Cloud to survive. AI to thrive: How CXOs are navigating the path to data-driven reinvention”

The Summit’s keynote featured Hari Sivaraman, head of AI content strategy at VentureBeat, in conversation with Accenture’s Sanjeev Vohra, global lead – applied intelligence.

Post-pandemic, there’s been a massive shift toward data, AI, and cloud to create greater good, greater revenue, and greater efficiencies.

Vohra identified four key fundamental changes he and his team have seen, particularly in the past year. First, is that cloud and data have come together as superpowers. On the one side, he explained, is the proliferation of cloud which provides much higher levels of compute power and the flexibility to scale up and scale down, depending on the need. That’s combined with vast amounts of data now available both inside companies or obtained from third-parties.

“Data and cloud are a huge trend we see powering the entire planet and it has really advanced during the pandemic,” he said.

The second trend is that the C-suite from companies across industries are now actually interested in these technologies and how they can be used to derive business value. “It has  has moved out of the experimentation zone, or pilot zone,” he said, “to be used for scale.”

Speed is the third trend. As Vohra explained, “Nobody wants to spend two years, three years trying to drive value. [Business leaders] are really getting serious about saying what can be done in six months.”

The last trend is talent. It’s scarce, and the demand is coming from everywhere. So companies now are having to make important decisions about how much investment is required for building staff, and what portion is focused on building internally versus recruiting from the outside.

Later in the discussion, Vohra shared one of the projects he is particularly excited about. Along with Intel and the Philippines-based Sulubaaï Environmental Foundation, Accenture is saving the coral reef with AI and edge computing that monitors, characterizes, and analyzes coral reef resiliency. Accenture’s Applied Intelligence Video Analytics Services Platform (VASP) detects and classifies marine life, and the data is then sent to a surface dashboard. With analytics and trends in real-time, researchers make data-driven decisions that are helping the reef progress even as we speak (or as you read).

Cigna C-suite executives discuss the impact of AI and digital interactions in transforming the health of their customers

During the AI in health panel, Gina Papush, global chief data and analytics officer at Evernorth/Cigna, had a conversation with Joe Depa, global managing director at Accenture, about how they’re using actionable intelligence to make health care more predictable, efficient, and most importantly, effective.

Their major focus over the past year and a half has been been understanding the impact of COVID geographically and across different population segments.

“One of the things we’ve uncovered is that without a doubt, there are differences in terms of how COVID is impacting different groups of customers, and particularly Black and Hispanic customers,” she said.

The organization partnered with their clinical and customer experience teams, working with employers locally in those markets, to bring a concerted, data-driven efforts to drive outreach. They proactively distributed PPE and education about preventing infections, and managing illness. And as vaccinations rolled out, they worked with customer employers to get these to vaccination sites.

Once they shifted focus to studying post-COVID effects, particularly long-haul COVID, they found that in patients with long-haul COVID, many customers have pre-existing chronic conditions such as heart inflammation and heart disease, which are prevalent at higher rates in communities of color. Now they’re focused on identifying risks, and data science teams are building models and applying models to identify those who may be at risk post-COVID for severe complications.

“It’s critical that post-COVID care continues, and our predictive analytics enable us to be more pinpointed in driving that care to the right folks,” Papush said.

Understanding consumer behaviour with big data & delivering AI powered products that offer personalized recommendations

This AI in retail panel, presented by Accenture, unpacked the ultra-personalization trend with AI leaders from DoorDash, Nike, and Accenture.

“It became more apparent every day that the post-pandemic acceleration of digitization has changed the way people consume and interact with products in all categories,” said Lan Guan, applied intelligence global solutions AI lead at Accenture. “AI has leapfrogged to indulge consumer demand for exactly what they want, when they still want it. That’s what ultra-personalization is all about.”

For DoorDash, this personalization centers around what the company calls “the restaurant selection problem,” explained Alok Gupta, head of data science and machine learning at DoorDash.

Consumers come to DoorDash with a specific food in mind. Their data scientists are focused on understanding what that desire is, and identify potential new restaurant partners that can help make the DoorDash app’s restaurant and food selection as robust as possible.

With digital demand exploding at Nike, their whole model had to shift, said Emily White, Nike VP of enterprise data and analytics. The company used AI and machine learning to automate internal processes to gain speed and launch a new distribution facility to fully support their growing digital demand.

Her team created a replenishment engine to read the signal, identify available inventory across all of Nike’s distribution centers and stores, and determine which products should be allocated to the Adapt facility in Tennessee to best serve the southeast region. It’s their largest distribution center worldwide, built to distribute the company’s Nike and Jordan products to individual consumers, wholesale customers, and Nike’s retail channels as efficiently as possible in the new digital-first world.

“The outcome of this is decreasing transportation time and cost, improving our sustainability, and helping us react faster to our local demand,” she said.

One of Accenture’s clients, a fashion brand, used AI and an ultra-personalization approach to go from passively offering just a few clothing collections a year to responding to what’s still hot in the market. They collect real-time consumer feedback from across social media platforms with AI and machine learning. Within just a couple of hours, designers translate this information into product ideas and send them to micro-studios for experimental production.

“Two quick results here,” Guan said. “25% growth in yearly revenue, and 29%-plus increase in revenue-per-visit, all because of that ultra-personalization.”


About Joyk


Aggregate valuable and interesting links.
Joyk means Joy of geeK