My papers

This is my final papers about Personalized Seacrh

Here is my class presentation

Week - 11 Resource

Robust Collaborative Filtering is a new filtering concept that uses Robust statistics. Robust statistics is an area within statistics where estimation methods have been developed that detoriate more gracefully in the presence of unmodeled noise and slight departures from modeling assumptions. This prevents against noisy ratings that might affect the recommendation results and also different attacks (e.g. profile injection attacks, shilling attacks) through the use of M-estimators and Robust Matrix factorization. But the experimental results from this paper show that application of M-estimators does not add significant stability; modified SVD algorithms outperform RMF in robustness. However the presented algorithm can outperform existing recommendation algorithms in recommendation quality.

Week - 11 Pandora

- First by logging, “It was not easy” by Cece Winans was played from my “Cece Winans” station. I noticed that Pandora always plays the first song from the main artist in the station. I tried to find people with similar profiles as mine, but Pandora doesn’t provide this kind of features. You search for person, music station by names; Pandora should provide an option like “See matching profiles.” In here people should browse through similar users and find other interesting things. Because people expectations are malleable, unstable and vary. It might return a high degree of serendipity.

- I noticed also that Pandora did not provide a high degree of privacy from the profiles I glanced through. I don’t know if it’s users’ preferences, because it is easy to detect and locate users. Pandora should provide an option for anonymized profile.

- On more weird thing, Pandora returns songs that I’ve already rated, but unrated. I have to rate them for a second time !!!!!!. Overall my “Cece Winans” station is almost accurate.

Week - 10 Resource

Personalization: Reducing Complexity for easier, more intuitive User experience.

As digital content, and communication become richer and complex, it is indispensable to personalized user experience in mobile platform. The approach Motorola is taking is personalization. It places user in the center of interaction, application and computing activities. A vision that Motorola had since the 90's since when they initiated many research projects and developed many tools. Motorola sees three personalization emerging: user interaction personalization, content personalization, and applicatios and services personalization. Among initiatives and platforms with personalization as core concept are: Motorola' s intelligent user interaction framework, Seamless mobility initiative, Motorola SCREEN3, MobiLife platform. Motorola is also trying to develop personalization standards by working with many other groups.

Week - 9 Pandora

- Pandora is getting closer and closer to my preferences in my Donnie McClurkin station. For instance, by loggging today, i got "Great and Mighty is our God" by Donnie McClurkin. I decided to perfect my Psalms station, the first song was "Peace,Love & Understanding" by Robben Ford; this song don't reflect my prefences. My next suggestion was "Can't give up now" by Mary Mary, since i added this artist in this station, i think Pandora picked it up from my profile whick is ok. My next song was "Purify" by Cece Winans.

- Pandora provides listeners with an option "Quick mix" where different stations can be mixed. Listeners can also chose music genre. It also recognizes stations based on their music genre they contain. Curiously, Pandora recognizes my Donnie McClurkin and Psalms stations as rock station. This is not true for my Donnie McClurkin, 'cause in this station i mostly flagged and liked songs related to Donnie McClurkin (is Donnie music rock?). For the Psalms station, Pandora randomnly flshes my rock and folk music though i'm trying to input as much as i can my preferences.

-

Week - 9 Resource

In this article , writer jemina Kiss argued that the speculated Web 3.0 will be about personalization. The site held a constest asking readers for their web 3.0 definitions, the favorite definition came from Robert O'Brien, Web 3.0 is defined as "decentralized asynchronous me". With the promise of semantic web, machines understand things in human terms, and can apply that knowledge to your attention data. The result we'll have a web that knows what we want and when we want it. So it is all about personalization and recommendation. Just a speculation from Jemina !!!!!! What's your idea ......

Week 8 - Resource

PRES is a recommender system that recommends links (hyperlinks) based on content-based filtering. A content-based filtering system selects items based on the correlation between the content of the items and the user’s preferences. PRES makes recommendations by
comparing a user profile with the content of each document in the collection. Term weights of document can be determined by using the tf-idf scheme. The user profile is represented with the same terms and built up by analyzing the content of documents that the user found interesting. Finally similarity measures (cosine) are used to measure how similar is a document to user profile and those (links) unred are recommended to the user.

Week - 8 Pandora

- Pandora started my session by playing “Church medley” by Donnie McClurkin and when I clicked “Now playing” section in the bottom of the page, a short biography of the singer is described. From the section, Pandora also lists similar artists related to Donnie McClurkin. Next “Perfect Peace” by Marvin Sapp was played, curiously “similar artists” list of this singer is closely related to that of Donnie McClurkin. From this observation, we can deduce that Pandora is playing from similar artists to Donnie McClurkin.
- “What is new” section presents new products in Pandora store. Pandora mobile allows you to listen to your music stations or create new ones from your mobile device (cell phone). Pandora in the Home partnered with the leading makers of home digital music systems to provide the full Pandora listening experience without the need for a PC to even be turned on.
- Pandora got a little close to my desires for my Donnie McClurkin station that I almost enjoy every song (NO NEED to track songs). With Pandora also you can choose stations by genre. It will provide you different music genre with their subcategories that you can choose from.
- “Pandora Presents” section announces festivals, events, or activities conducted by Pandora team. On march 28, Pandora will hold Great venues : SXSW, part 1 event to present and promote young bands.

Week - 7 Pandora

- "Trouble ain't" by Mary Mary was playing on my login. The song was nice and i liked it. I was a little curious to know about what Pandora is doing and clicked "Your Profile" tab. In the middle frame, there is a page section "About me", here Pandora gives an option to write something about myself. Below comes "Bookmarked songs" page section, here i can store songs that i heard on Pandora. Next comes "Bookmarked Artists" section where i can store artists that i heard on Pandora.

- On the right panel, "Bookmarked People" to store other listeners that i found on Pandora. I can add people by providing their emails to the textbox.

- Curiously, Pandora stopped play weird song that i didn't like and played songs that somehow i liked. Now, "Broken but i'm healed" by Byron Cage was aired live and the song was enjoyeable. "Be blessed" by Yolanda Adams, "You are" by Kirk Franklin were proposed to me and i really enjoyed them all.

- Sudently Pandora started weird songs that i didn't like before. For instace "Legion of deads", "Are you sure" (especially i remeber not to like this song and notified Pandora). Interesting !!!!!

Week - 7 Resource

Oracle Database 10g provides an integrated personalization and recommeder engine.OracleAS Personalization is a part of Oracle Application Server 10g, Enterprise edition,providing real-time personalization for e-business sales channels, such as Web Stores, application hosting environments and call centers. It uses Transactional naive Bayes and predictive association rules data mining algorithms to find hidden patterns and customer profiles that drive personalized recommendation. OracleAS personalization has an API that allows e-business to offer real-time personalizations to their registerd customers and Web visitors for any Java web site running on OracleAS. It also provides session tracking (current vs historical) at the API level to provide fine-tuned recommendations for each visitor.

Week - 7 Pandora

- Pandora played “All I need” by Cece Winans, a song that I already rated as favorite. But my purpose today is to better my Psalms station, since it is not detecting my tastes. I will more interact and feedback with this station till it will hit one of my likes.

- “River is waiting” by John Forgety played and it flagged it as “disliked it”. Next I got “What a man” by Soul summit, I didn’t like. I noticed that Pandora was not even coming close to my likes range; I decided to add more music types to this station.

- Using Pandora “Add more music to this station”, I added Mary Mary singers in my station’s list. Pandora directly detected the changes and played “My desire” by Kirk Franklin (although I don’t like him very much), music type that this station was not able to play a while ago.



- Next “Good to me” by Mary Mary was played, Pandora got closer to my choices. My real desire is to test Pandora if it would be able to pick Don Moen, Donnie McClurkin from this station without explicitly entering them in my list but from my choices.

Week - 6 Pandora

- Pandora played 'Faith is the key' by Alvin Slaughter, i like it. I decided to add a little variety to my tastes by creating an central african gospel station. Unfortunately, Pandora's inventory was not that very large.

Week - 6 Pandora

- By now Pandora somehow figures out that I like Donnie McClurkin too much, and proposes me one of his hit most of the time I log in. Today I got Victory Chant by Donnie McClurkin (from my McClurkin station). Next ‘Yes Lord’ by Andre Crouch was played, this name sounds familiar to me but I don’t remember hearing one of his songs. I was curious to hear this song till the end. The song had an average raking according my taste, I flagged it as ‘I liked it’. Next i got 'Make me over' by Tonex, it's an artist that i don't really like much, 'i didn't like'.

- 'I need Thee every hour' by Don Moen was played, from one of my favorite album 'Thank you lord'. Next 'Breath' by John Tesh, a song that i've been look for but i didn't know who sang it, i got it !!!

The implementation of knowledge-based recommender system

The mainstream of recommender systems rely on collaborative or content-based filtering. This paper proposes a knowledge -based recommender system that interacts with the user and matches his/her requirements based on knowledge of the product domain using an inference engine (deduction based on rules). The knowledge base of the product and inference engine is implemented by JESS and java servlet. This approach requires a domain expert to model the object to macth with user requirements. After finding candidate products, cosine is used to measure similarity between the requirements and the products' features. The paper presents a sample implementation of a cellular phone recommender that uses phone model.
Click here to read the paper

Week 5 - Pandora

- Today my Cece Winans station played ‘Jesus, You are beautiful’ by Cece Winams, an album that a friend let me borrow one time which I liked. Next I got Worship medley from Dr. Bonnie Hunter, I didn’t like it very much. But from this station, Pandora was getting a little closer to what I like. ‘You are loved’ Cece Winans was played next, it was my first time to hear to song but I liked it.

- ‘The heart of the matter’ by India Arie was played, a genre that I didn’t like to be played on this station. I asked Pandora why the song was played; it replied that this song matches my profile since it is a new style of R&B with strong vocal. I don’t think that India Arie is in the category than Cece Winans or Yolanda Adams.

Week 4 - Resource

Recommender systems recommend suggestions from data coming from s single organization database. Thus, they improve their accuracy with data availability and increased user's interactions. Systems that don't have data can't accurately predict. To solve the cold start problem, different recommender systems can share their knowledge in order to better the recommendations.


This paper can be viewed from IEEE site using the library site.

Week 4 - Pandora

- Now it's playing 'The awsome God' by Kevin Vasser, and it's one of my favorite. I reached my limit of song/per hour per station. Pandora suggested me to create another station if i wanted to continue listening to more music.
- I created a new station with Cece Winams as my favorite artist. First the station played 'All that i need' by Cece winams. Next it palyed 'All of my life' by Trin-I-Tee 5:7. It didn't like this song.
- Next Pandora played 'Fragile Heart' by Yollanda Adams, i really like it. By the way , i was lookig for this song on the net.

Week 4-Pandora

- Today Pandora played from my PSALM station "I'll be" by Little Milton; it's way off to my tastes. Next it played 'No one special' by the Robert Cray band, this one is alo off.
- I asked 'Why this song was playing' , pandora answered that based on what i told them so far. That the song features new R&B styling, a subtle vocal, mild rythmic, major key tonality and electric guitar. It is a little bit close to what i like but not this kind of music.
- Midnight to Day (by Tommy Castro) was played, but it's not even close to what i like. Curiously, Pandora starts playing 80's rythmic and vocal type music. (not me !). 'Magnificient Sanctuary Band' by Susan Tedeschi was next played, i didn't like 'cause it's not my kind. Next 'come around' by Needle was playing, it was clear to me that Pandora don't understand my taste for this station.

- I decided to to change station. I went back to my Donnie McClurkin station. Pandora played "The Potter's House" by Tramaine Hawkins, i sonehow liked it (half and half satisfaction)
- 'I will sing' by Donnie McClurkin was played and i like it.

Week 3 - Pandora

- I created a new station called "Psalm". I didn't like the first song (Good news by Michelle). I didn't like the next song "His precious Love" by Otis Clay.

- Next i got "Midnight train to Georgia" by Joan Osborne (what!!!!!!!!!). way off!
- Pandora starts returning all the artists that i previously turned down but with different song.

- As turned down so many songs, Pandora warned me that i exceeded the number of songs that i'm allowed to hear per hour

Week 3 - Pandora

- Today Pandora played Caribean Medley (Donnie McClurkin). Next I've delivered (by Georgia Mass Choir), i didn't like it too much
- I need Thee (by Don Moen) was played, i like it. You know me more (By Hillsong). Curiusly Pandora played some kind of music that i like.

Week 3 - Resource

Recommender systems use a database about user preferences to predict additional topic or products. In order to achieve this, conventional recommender systems rely on user x item informations and ignoring other factors that influence user preferences. By including temporal data, recommender system will efficiently make recommendation.

(The paper can be view by login the IEEE through the Library site)

Pandora system

- Upon logging Pandora played: All to You/I surrender (by Don Moen) an album that like and owns. I requested Pandora to play similar music.

- God still heals by Dawayne Wood, was payed next. A song that i didn't like too much, then Pandora played When it's all been Said and Done by Don Moen , but forwarded to next station in order to discover my recommendations.

- I got to station that was playing I just want you (by Vicki Yohe), i didnt like and I understand by Smolie Norful, i gave thums up. Next i got victory Chant by Donnie McClurkin that i really liked...............

Week 2- Pandora

- "Oh The glory of his presence" by Carlton Pearson on Azusa Praise jubilee album was played. Pandora detected that i might like Praise songs. I didn't react to it, next "Come Holy Spirit" by Micah Stampley (Album: The songbook of Micah). I pointed that I liked it. The next song was "We lift your Name" by O’Neal and Expression of Praise. I responded that i didn't like and the song interrupted then it palyed "Great is that faithfullnes" by Troy Sneed on State of Worship album

- Pandora played "The awesome God" by Kevin Vasser (Album:So Glad) and really liked the song. In the meantime, on the menu i requested that i wanted to add more kinds of music to this station and added Don Moen.

- Great is that faithfullness was abruptly changed and played "When it’All been Said and done" by Don Moen. I really liked the song.

- Next Pandora was really off target by playing 'You are' by Harriet Schock (on American Romance album)

Week 2- Pandora

-By signing in my account, Pandora played "He is the greatest" by john P. Kee and the New Life. I flagged it as " I didn't like it and don't play it for a month". Pandora acknowledged my request.

-Next "Jesus, the mention of your name" (by Donnie McClurkin) was played from a compilaton of artists, the album was: on Praise and worship : all star Gospel hits).
On the menu i answered the question 'why did you palyed this song?' I clicked About the Author.

News - Week 2 -Jester: The Online joke Recommender

Jester is online system that recommends jokes based on previous joke ratings. It uses a collaborative filtering algorithm called Eigentaste.

Eigentaste uses universal queries to elicit real-value ratings and applies principal component analysis to facilitate dimensionality reduction for users' clustering. It is proven to be more efficient than nearest-neighbor techniques and other techniques based on output data from JESTER.

News -Week 1

link: [Leaving the era of search and entering one of discovery] :

A system able to discover user personality, feeling, emotions through interactions and thus deciphering user's preferences and dislikes. Discovery is when something you like finds you! The system knows more about you then you know about yourself.

Metrics

After setting the account, Pandora asks for a sample song. Pandora starts by playing the station that examplifies the kind of music i chose.
My first choice was Donnie McClurkin, the system found a station that was playing "Days of Elijah" (Donnie McClurink). Under the main song' window there is a "Guide us" tab that asks user to rank the song :

- I really like this song ; play more like it (1)
- I don't like this song; it's not what my station should play (2)
- I'm tired of this song; don't play it for a month (3)
- Why is this song playing? (4)
- I want to add more kinds of music to this station (5)

I didn't rank the song (Days of Elijah), next "He's able" (by Kirk Franklin). I don't really like Kirk Franklin but liked this song. I chose (1) for both songs

My first entry

First of all, i registered to pandora by creating an account. I provided a fake profile and wrote it on my notebook in order to track how pandora will get closer to the profile