Jysk Fynske Medier
About Jysk Fynske Medier
The work made for Jysk Fynske Medier runs to 1 verified project — mostly content, DDA2022 and Strategy.
Inside the Jysk Fynske Medier network
The people, teams, and projects that make up the Jysk Fynske Medier network
Featured project
Personalization of Content Across Channels
STEP has developed an algorithm for personalizing content for Fyens Stiftstidende (and in the long run the rest of Jysk Fynske Medier's regional dailies). The algorithm is built centrally in our data setup and is therefore independent of platform. This means that regardless of whether you receive a newsletter or go to fyens.dk, the content is personalized so that it matches your interests and your reading patterns. In other words: You are affected by content that is relevant to you in particular - no matter where you meet Fyens Stiftstidende. Danish Hvordan får man distribueret det rigtige indhold ud til de rigtige mennesker? Der findes uendelige muligheder for at udgive og distribuere information og indhold i dagens Mediedanmark. Og aviser har i mange år haft det svært. De bruger mange ressourcer på at producere en stor mængde indhold, mens mængden af tid, de har til at fange brugernes opmærksomhed, er ultrakort. For brugerne søger hen, hvor de får den bedste dækning af emner, de interesserer sig for. Og det gælder alle platforme. Så hvordan kan man som medie bedst muligt tilbyde præcis det indhold, som hver enkelt medieforbruger er ude efter? Hvordan kan vi skræddersy vores indhold til forbrugeren, så forbrugeren bliver præsenteret for præcis det indhold, han/hun er ude efter? Og hvordan kan vi gøre det på tværs af kanaler? Målet har været at få Fyens Stiftstidendes læsere til at konsumere mere og bruge længere tid på indholdet, fordi de blev præsenteret for indhold specifikt udvalgt til lige præcis dem. One size doesn't fit all STEP har udviklet en algoritme til personalisering af indhold for Fyens Stiftstidende (og på sigt resten af Jysk Fynske Mediers regionale dagblade). Algoritmen er bygget centralt i vores data setup og er derfor uafhængig af platform. Det betyder, at uanset om du modtager et nyhedsbrev eller går ind på fyens.dk, så er indholdet personaliseret, så det matcher dine interesser og dine læsemønstre. Altså: Du rammes af indhold, der er relevant for lige netop dig – uanset hvor du møder Fyens Stiftstidende. STEP har skabt en algoritme, der benytter collaborative filtering. Det er en strategi, hvor algoritmen har fokus på brugernes adfærd i stedet for at kigge på parametre som kategori, forfatter eller geografi (den tilgang, der kaldes content-based filtering). Det betyder, at vi har trukket data om læsemønstre baseret på artikel-ID’er og 1. parts bruger-ID’er fra vores CDP (Customer Data Platform). Løsningen kan faktisk også matche indhold til brugere, vi endnu ikke har adfærd på, hvilket ellers er en kendt begrænsning ved brug af collaborative filtering – det, som også er kendt som cold start. Algoritmen baserer derfor anbefalingerne på baggrund af andre med lignende læsemønstre. Det er en løsning, der kræver mindre vedligeholdelse, idet teknologien automatisk opdager mønstre. Alle manuelle ressourcer er sparet væk, da løsningen er 100 % automatiseret og drevet af data. Den traditionelle content-baserede løsning er begrænset af, at maskinen tror, at fordi du én gang interesserede dig for X, så vil du altid gøre det. Men faktum er, at mennesker ændrer sig, og det kan denne løsning bedre opfange, fordi den er baseret på, hvad andre, der ligner dig, gør. Algoritmen indeholder et sæt regler, som den bruger til at definere, hvorvidt folk skal eksponeres for personaliseret indhold eller ej. Brugeren skal have læst mindst otte artikler, ligesom brugeren skal have været aktiv inden for de sidste fire uger. Derudover er der kortere aktualitet for 112-artikler, ligesom brugeren naturligvis ikke får anbefalet artikler, han/hun allerede har læst. Surprise. Det virker at være personlig. Det personaliserede indhold blev implementeret på fyens.dk og i Fyens Stiftstidendes nyhedsbreve i sommeren 2021. Herefter blev en A/B-test igangsat med fokus på nyhedsbrevene. En halvdel modtog nyhedsbreve, som ikke benyttede sig af personalificeringsmodellen, mens en anden halvdel modtog nyhedsbreve med personligt anbefalede nyhedsartikler foreslået af STEPs algoritme. Resultatet af undersøgelsen var entydigt. Click raten er mere end 40 % højere i de personalificerede nyhedsbreve end i de standardiserede. Helt præcist klikkede 52.150 personer ind på fyens.dk gennem en personlig henvisning kontra 36.607 i standardnyhedsbrevet. Dermed skaber algoritmen mere trafik til sitet, hvor vi også har registreret, at læserne har kvitteret med længere tid på sitet og klik på de anbefalede artikler. Derudover har der været en reducering på 50 % af permission-churn i den gruppe, der modtog AI-indhold i nyhedsbrevene. Løsningen er 100 % automatiseret, datadrevet og kræver ikke manuelle ressourcer i redaktionen. Det har frigivet ressorcer til andre redaktionelle opgaver. English (Google translate) How do you get the right content distributed to the right people? There are endless possibilities for publishing and distributing information and content in today's Media Denmark. And newspapers have had a hard time for many years. They spend a lot of resources on producing a large amount of content, while the amount of time they have to capture users' attention is ultra-short. For users to go where they get the best coverage of topics they are interested in. And that applies to all platforms. So how can you as a media best offer exactly the content that every single media consumer is looking for? How can we tailor our content to the consumer so that the consumer is presented with exactly the content he / she is looking for? And how can we do that across channels? The goal has been to get Fyens Stiftstidende's readers to consume more and spend more time on the content, because they were presented with content specifically selected for exactly them. One size does not fit all STEP has developed an algorithm for personalizing content for Fyens Stiftstidende (and in the long run the rest of Jysk Fynske Medier's regional dailies). The algorithm is built centrally in our data setup and is therefore independent of platform. This means that regardless of whether you receive a newsletter or go to fyens.dk, the content is personalized so that it matches your interests and your reading patterns. In other words: You are affected by content that is relevant to you in particular - no matter where you meet Fyens Stiftstidende. STEP has created an algorithm that uses collaborative filtering. It is a strategy where the algorithm focuses on the behavior of the users instead of looking at parameters like category, author or geography (the approach called content-based filtering). This means that we have extracted data on reading patterns based on article IDs and 1st party user IDs from our CDP (Customer Data Platform). The solution can actually also match content to users we do not yet have behavior on, which is otherwise a known limitation when using collaborative filtering - what is also known as cold start. The algorithm therefore bases the recommendations on the background of others with similar reading patterns. It is a solution that requires less maintenance, as the technology automatically detects patterns. All manual resources are saved as the solution is 100% automated and data driven. The traditional content-based solution is limited by the fact that the machine thinks that because you were once interested in X, you will always do it. But the fact is, people are changing, and this solution can better capture that because it's based on what others like you do. The algorithm contains a set of rules that it uses to define whether people should be exposed to personalized content or not. The user must have read at least eight articles, just as the user must have been active within the last four weeks. In addition, there is a shorter timeliness for 112 articles, just as the user naturally does not get recommended articles he / she has already read. Surprise. It seems to be personal. The personalized content was implemented on fyens.dk and in Fyens Stiftstidende's newsletters in the summer of 2021. After this, an A / B test was initiated with a focus on the newsletters. One half received newsletters that did not use the personalization model, while another half received newsletters with personally recommended newsletters suggested by STEP's algorithm. The result of the study was unequivocal. The click rate is more than 40% higher in the personalized newsletters than in the standardized ones. Exactly 52,150 people clicked on fyens.dk through a personal reference versus 36,607 in the standard newsletter. Thus, the algorithm creates more traffic to the site, where we have also registered that readers have signed up for longer time on the site and click on the recommended articles. In addition, there has been a 50% reduction in permission churn in the group that received AI content in the newsletters. The solution is 100% automated, data-driven and does not require manual resources in the editorial office. It has released resources for other editorial tasks.