Artificial Intelligence Marketing Lab

How to filter out isolated nodes or small communities

aimarketinglabprof — Thu, 03 Oct 2019 00:47:36 +0000

Very often, the twitter network data we collected has many isolated nodes or isolated small communities. To get a meaningful network visualization graph, we may want to filter out those nodes or small communities in Gephi. We next give a step by step instruction for finishing this task.

First, please calculate modularity class in the statistics tab.

Next, we need choose to use modularity class to do filtering under the tab of “Filter”

After we clicked Filter button, you can see that there are only 1788 of 4555 nodes left. We can save and work on this filtered data.

You can also export the filtered network data by following the following steps:

By clicking Workspace 2 tab, you now work in Workspace 2 ( the exported filtered data). You can then export the date in this workspace to a graphml file.

Collecting Twitter Network Data with R

aimarketinglabprof — Tue, 26 Sep 2017 19:05:37 +0000

R is a free software environment for statistical computing and graphics. You can download it from its official website. https://www.r-project.org/

In the following, we share the R codes for collecting twitter search network data. This network data is created from tweets containing #BoilerUp hashtag. We create a node for each unique Twitter user who either tweeted one of those tweets, was replied to in one of those tweets, or was mentioned in one of those tweets. These nodes were connected with edges of tweeting, replying, and mentioning. If a user X1 was replied or mentioned by another user X2, the edge between X1 and X2 represents a one way connection from X2 to X1.

First step, you need install package twitterR, igraph, reshape, reshape2 from R. The package graphTweets is not offered within R and one need download it from https://cran.r-project.org/web/packages/graphTweets/index.html to your computer. Then you can install this package from R.

Second step, you need gain the authorization from twitter for your R application. Please read the article here to understand the detailed process. http://thinktostart.com/twitter-authentification-with-r/

Third step, run your codes in R.

library(twitteR)
library(igraph)
library(graphTweets)

#setup_twitter_oauth() function uses the httr package
# replace these value from your register on twitter
#Twitter Authentication with R

Consumer_key=”YOUR API KEY”
Consumer_secret=”YOUR API SECRET”
access_token=”YOUR ACCESS TOKEN”
access_token_secret=”YOUR ACCESS TOKEN SECRET”
setup_twitter_oauth(Consumer_key,Consumer_secret,access_token,access_token_secret)

#pull and parse tweets
tweets=searchTwitter(“#BoilerUp”, n=5000, lang=”en”)
tweets=twListToDF(tweets)
edges=getEdges(data = tweets, tweets = “text”, source = “screenName”)
nodes=getNodes(edges)
graph=graph.data.frame(edges[,1:2], directed=TRUE, vertices=nodes)
write.graph(graph, “D:/SNAdata/boilerup.graphml”, format=”graphml”)

More information about twitterR package can be found from the following link.
# https://cran.r-project.org/web/packages/twitteR/twitteR.pdf

Having Trouble Launching Gephi: Gephi and Java

aimarketinglabprof — Tue, 26 Sep 2017 18:35:15 +0000

Gephi is a free software which costs users nothing. On the other hand, we may not expect the same level of stability as provided by the commercial software. One of the common problem is the comparability of Gephi with the Java platform installed.

Gephi 0.9.0, Gephi 0.9.1, and Gephi 0.9.2 have been tested on Java 7 and Java 8. Therefore, you are recommended to have these versions of Java installed on your computer. Check the following website for these versions of Java platforms:

http://www.oracle.com/technetwork/java/javase/downloads/index.html

If you have Java installed, but the Gephi reports that it cannot find the Java platform. One possible reason is that the Gephi installation failed to locate the path of your installed Java platform. You can correct this problem by following these steps:

(1) Open configuration file under the installed folder of your Gephi, for example, C:\Program Files (x86)\Gephi-0.8.2\etc\gephi.conf, depending where you install your Gephi.

(2) Specify jdkhome=”C:\Program Files\Java\jre8″, the specific path you put here depends on where you install your Java platform. Please de-comment out this line by removing the “#” symbol after you change the default specification.

Visualize a network data in Gephi step by step

aimarketinglabprof — Sun, 27 Aug 2017 20:22:36 +0000

Visualization of social network data has provided an interesting approach to understand the diffusion of consumer information among customer network. In the following, I provide a step by step instruction of visualizing Nike’ s famous hash tag: #justdoit. I first collect all recent tweets that contain the hash tag of “justdoit” from Twitter. In the data, each twitter account that mentions the hash tag is treated as a node. If the tweet is a reply to another tweet or a mention of another tweet, an edge is added between the account and the other account. With the collected the data, we analyze the network and make a graph of the social network of reply/mention of hash tag “justdoit” in another free software, Gephi. You can enlarge the pictures in the following by click them.

(1) Open the data file in Gephi.

(2) Calculate statistics and size the nodes based on a measurement of centrality.

(3) Partition the network into several communities based on the calculated modularity.

(4) Run a layout algorithm to get a better layout of nodes and edges.

(5) Next, show the labels for nodes and run a label adjust algorithm.

(6) To preview the graph, switch to the “Preview” window. To show the label, you need check the “Show Labels” box.

(7) In Data Laboratory window, you can sort the nodes based on some centrality measurement.

Let’s Talk Programmatic: Marketer Spotlight – Dr. Qiang Liu

aimarketinglabprof — Tue, 27 Jun 2017 20:23:47 +0000

Dr. Qiang Liu is fascinated with the world of machine learning and its effects on the future of advertising.

He’s an Assistant Professor of Management at Purdue’s Krannert School of Management and a researcher with a highly analytical background. Liu earned a double major in Information Management and Economics at Peking University and has advanced degrees from UC Berkeley and Cornell in Statistics and Marketing.

His research and teaching focuses on modeling consumer behaviors, competitive strategies, pharmaceutical marketing and social media marketing. After hearing his excitement about the future of advertising, we just had to share his insights with you:

Programmatic Advertising.org: Thanks for your time Professor Liu. I know you’re a very busy man! In your opinion, how is programmatic advertising – the growing prevalence of marketing automation, data mining and the cloud – changing the industry?

Dr. Qiang Liu: Programmatic advertising increases the effectiveness and efficiency of advertising with a real-time system and rich data from multiple sources. It facilitates the optimization of communication between brand managers and potential individual consumers. With programmatic advertising, brand managers can reach the right customers, with the right message within the right context at a lower cost. It is disrupting the traditional advertising industry, such as traditional media e.g., print and TV. It is also disrupting traditional advertising agencies. The firms without digital advertising expertise are definitely impacted the most.

Programmatic Advertising.org: Have you seen any digital marketing campaigns that you especially like? Why?

Dr. Qiang Liu: Yes, Oreo’s brilliant Super Bowl blackout tweet really impressed me. The quick reaction to the Big Game blackout highlights the importance of a “real-time” marketing campaign in a digital age.

Programmatic Advertising.org: That was an incredibly fast turnaround time that definitely went down in history. What excites you about the future of digital advertising? What are you looking forward to?

Dr. Qiang Liu: The big data and deep machine learning developments excite me the most about the future of digital advertising. With this huge amount of data collected and the ability to learn automatically, firms can gain a more accurate, real-time, and personalized understanding of consumer behaviors and be able to give each consumer what they want at the right time and place.

Programmatic Advertising.org: I heard your excitement when you mentioned deep learning. I am very interested in hearing what you have to say on what you think the future of advertising will be.

Dr. Qiang Liu: The future of advertising should be more personalized and data driven. It should also be more subtle and intuitive. If you can make your advertising with data and at the same time keep creativity alive, you can tell the story that will probably work the best. This automatic storytelling has 2 dimensions. One dimension is in programmatic advertising – which is doing pretty well at being data driven automation. At the same time, you don’t want to lose that other side, the being more creative – the telling of an interesting story. How do we do that in the future?

That is something deep learning will do for marketing in the future. You’re not just getting pre-designed information and doing targeting with machine learning. You can actually createsomething, not just repeat something we’ve already seen before. With algorithms and rich data you can create some story that can be attractive to a particular individual consumer. In this sense, the compilation of big data with deep learning will dramatically change the landscape for digital advertising.

Programmatic Advertising.org: Can you clarify – what exactly is deep learning?

Dr. Qiang Liu: Deep learning is basically a more sophisticated machine learning. When it is applied in marketing, it’s not ad buying automation, it’s more than that. Here is an example.

You have so many tweets and so many Facebook posts, and with computer programming you can learn the sentiment of these comments. That is just the beginning of deep learning. Deep learning became even smarter.

Let’s say you are on Facebook and you are wondering how it knows your friend’s faces. This software that recognizes what people look like is deep learning. You can do some amazing things with it.

I can share with you another example of deep learning that was given by my colleague, Professor Xiao Wang at Purdue statistics department. Deep learning is even used in car anti-theft systems. Let’s say you store some anti-theft device or program with deep learning capability in your car. It will learn your driving habits, so if someone steals your car, the computer program will notice that the driving habits are now different from the owner. It will call the owner and text message them to let them know the car was stolen and where it is.

In summary, deep learning is a more sophisticated and abstract level of machine learning.

Programmatic Advertising.org: Deep learning seems to be very relevant to digital marketing. Wow – this really sounds like something from the future.

Dr. Qiang Liu: What we are doing now is just making copies of advertising manually. You then use the programmatic advertising mechanisms to sell this to a particular individual consumer and get the media place. But that is not enough.

With deep learning software, as I said earlier, we can actually create something new every time. We will find out through algorithms that there is a big idea you need to communicate with an individual consumer. The machine would have actually learned about the consumer using gender, age, past purchases, movie viewing, articles the consumer has read – everything – to make a mix of information and understand how to speak to this person. I think in the future we’ll see deep learning software creating the advertisements themselves.

We are not doing this now. Google and Facebook are already using deep learning for facial recognition, but we haven’t seen advertisements made with it yet. As I saw earlier in the Super Bowl blackout, Oreo had a team of social media marketers that watch events and when they thought they had a good opportunity, they decided let’s do it. It took them a little while to put together that campaign. With deep learning they could do this same process automatically – almost instantly.

This is something that really may happen.

Programmatic Advertising.org: So, if I understand all of this correctly, deep learning is just algorithms, programs and software. If it is creating our advertisements for us, will we even need creatives anymore? Marketers?

Dr. Qiang Liu: No, you still need people to make these things work. It takes time and effort to see something like what I have described to you.

Programmatic Advertising.org: What are some of your concerns about the future of digital advertising?

Dr. Qiang Liu: The privacy issue imposes a major challenge to digital advertising in the future. Legislators are currently introducing bills to protect consumers’ privacy, which is beneficial to the public in principle. However, overly protective actions may potentially make rich data unavailable to digital advertisers and prevent the industry from improving the effectiveness of targeting communication.

Programmatic Advertising.org: Thanks so much for your insight professor. This discussion has been amazing.

How to prepare for the Google Analytics Individual Qualification (IQ) exam?

aimarketinglabprof — Mon, 26 Jun 2017 14:30:35 +0000

The Google Analytics Individual Qualification (IQ) is a demonstration of proficiency in Google Analytics that is available to any individual who has passed the Google Analytics IQ exam. The Google Analytics certification remains current for 18 months from the date you pass the exam. This online exam contains 70 multiple choice questions and lasts no more than 90 minutes.

Steps for passing Google Analytics Individual Qualification Test

To prepare for Google Analytics Individual Qualification test, one should first go through Google’s online courses: https://analyticsacademy.withgoogle.com/course01/course . The course materials include instruction videos and reading articles. The assessment after each unit and the final assessment can be very helpful for you to review what you have learned. However, the assessment seems much harder than the real test. So please do not be deterred by those assessment questions.

Next, you can take advantage of an online study guide: https://support.google.com/partners/topic/6083717 .

Finally, you can find useful practice questions from this website : http://www.gaiq-center.com/exam-questions-answers/ These practice questions are very helpful, especially if you study the guide carefully.

To start for Google Analytics certification and AdWords certification exam, you should first join in Google partner. Please check this short video for the instructions of joining Google partner. https://www.youtube.com/watch?v=HNAB9yXoDYw

Good luck for the journey of passing Google Analytics IQ test and wish you be certified successfully.

Attention to Detailing: Would Reducing Pharmaceutical Sales Calls to Physicians Help or Harm Patients?

aimarketinglabprof — Sun, 25 Jun 2017 20:14:26 +0000

Attention to Detailing
Posted on November 17, 2015 by Claire Hall
Medical Sales

UConn Professor Asks: Would Reducing Pharmaceutical Sales Calls to Physicians Help or Harm Patients?

When a pharmaceutical company sends a representative to your doctor’s office to promote a new or existing medication, is that a benefit to you as a patient? Would restricting those visits bring greater fairness to the pharmaceutical industry—or prevent your doctor from being well-informed about treatment options?

Those are some of the questions that UConn marketing professor Hongju Liu and three colleagues tackled in a research paper titled, “An Empirical Model of Drug Detailing: Dynamic Competition and Policy Implications” that is pending publication in the journal Management Science.

Hongju Liu “The pharmaceutical industry is tremendously important and has recently undergone significant changes,” Liu said. “What we discovered could have ramifications for physicians, the industry and each of us as patients.”
-Hongju Liu

The vast amount of detailing—in-person calls by drug-company representatives—in the pharmaceutical industry has drawn interest from the public and legislators. In 2013, the Physician Payment Sunshine Act went into effect, creating transparency requirements resulting in physician practices and hospitals limiting pharmaceutical sales representatives’ access to doctors. Yet there has been little research about how these restrictions impact doctors’ prescription behavior and industry competition, Liu said.

Wide-ranging legislative proposals in different countries have been proposed to regulate and limit detailing activities, and there have also been efforts at self-regulation by both the industry and medical profession. The United Kingdom limits pharmaceutical firms’ sales promotion; Spain prohibits firms from providing more than 10 sample packages to a doctor in one year; and in 2009, Pharmaceutical Research and Manufacturers of America issued new guidelines on marketing to physicians, prohibiting non-educational and practice-related gifts, including pens.

In the U.S., sales representative access to physicians has become increasingly difficult. In 2013, some 45 percent of prescribers put significant restrictions on detailing representative access, compared with just 23 percent five years earlier.

For their research, Liu and his colleagues tracked the prescriptions for three popular statins, including Lipitor (Pfizer), Crestor (AstraZeneca) and Zocor (Merck). Statins are a group of drugs that reduce levels of fats, including triglycerides and cholesterol in the blood. In 2004, when the study began, statin sales surpassed $15.5 billion, making them the biggest-selling drugs in the U.S.

Liu and the other researchers studied 448 physicians who wrote almost 15,000 statin prescriptions and received more than 26,000 detailing visits. The study was based on data from 2002 to 2004 and mid-way through the study period, the new drug, Crestor, was introduced.

Liu and his colleagues wanted to know what the implications were for strategic behaviors as firms compete for the limited number of times they can meet physicians. Which firms gain, and which lose, in terms of revenue and profits? And what are the implications for the industry and public policy?

The researchers drew several conclusions:

By “leveling the playing field,” restrictive policies benefit firms that have the weakest detailing effects, while hurting firms with the strongest. These effects are especially pronounced on prescriptions for new drugs, which need to build up their physician interest. This may be undesirable when new medications are better therapeutically. Imposing a ceiling on detailing frequencies leads to a significant reduction in detailing across the board. A self-imposed restrictive detailing policy established by a single drug firm is not likely to succeed in reducing detailing levels of other firms. With firms’ detailing levels reduced by a restriction policy, the number of non-drug treatment or generic drug prescriptions by medical practitioners expands. But whether that is a benefit or hindrance to public health is undetermined.

Liu collaborated with Qiang Liu, a marketing professor at Purdue University, Sachin Gupta, a professor at Cornell University and Sriram Venkataraman, a professor at the University of North Carolina at Chapel Hill.

A/B testing with Google Analytics

aimarketinglabprof — Sun, 25 Jun 2017 18:34:42 +0000

A/B testing is a randomized experiment with two variants, A and B. Such experiments are commonly used in web development, email marketing, and advertising design. In web development case, you define a control or original page (page A) and an experiment or variation (page B) of the original page to test against. The goal of the test is to randomly expose your web traffic to the different versions of a page to determine which version will result in a better defined goal such as higher conversion rates, longer duration periods, lower bounce rate, and so on. The statistics behind A/B testing can be Z-test, t-test, or Chi-square test.

Google Analytics offers an easy to use tool (content experiment) for conducting A/B testing. The process of setting up an A/B testing in Google Analytics is kind of straightforward.

(1) In Google Analytics, go to Behavior-> Experiments

(2) Choose an experiment objective: this can be bounce rate, duration time, and conversion.

(3) Configure your experiment: set up the url for your original page and your experiment page.

(4) Paste the experiment code immediately after the opening head tag at the top of your original page. Also make sure both of your original page and experiment page contain Google Anlytics track code.

(5) Check and start the experiment.

Read the detailed instructions on setting up an experiment.