We are really happy to announce the AdHive Platform Beta release which will happen next week! It is one of the most important milestones of the project development so far.
AdHive dev team has been working on the Beta around the clock during the previous month and has designed many new features under the hood of our platform. The Beta version of the platform covers full cycle of launching advertising campaigns on YouTube and Instagram.
We have presented the platform’s Beta at Digital Summit Seattle. The product impressed the audience and initiated a lot of questions. Many digital marketing professionals from various fields (IT, media, entertainment and even fishing) were interested in our approach to influencer marketing and expressed their willingness to cooperate. Most of them said that they are looking forward to see AdHive Platform working in the USA.
In this article we cover the technologies used to build the Platform and functionality of advertiser’s and influencer’s Personal Accounts.
The Platform is a service that connects Advertisers with Influencers and provides them with handy tools necessary to run a successful campaign. In this regard Advertiser`s and Influencer`s Personal Accounts are the crucial parts of this mechanism.
Advertiser’s personal account
Upon registration on the platform advertisers get access to the personal account which is designed to launch native advertising campaigns through influencers.
Advertiser’s account has the following features:
- starting an advertising campaign
- checking previous campaigns reports
- uploading content to train the AI (pictures, words, stop-words)
- depositing and withdrawing ADH
- specifying user settings
To launch a campaign an advertiser makes just a few steps:
- register an account
- deposit ADH
- train the AI to recognise his materials (logo, product, words etc.)
- create a campaign (make a description, provide specific tasks, target audience and allocate the budget)
For the needs of advertisers a special deep targeting system has been developed. It allows the advertisers to filter target audience by age, language, sex, location and other criteria. The required information is provided by APIs of Instagram and YouTube and enriched with data collected by the STAT module.
Moreover, this type of personal account features an analytics dashboard with reports on their campaigns. Such reports include detailed influencers coverage, number of views, different audience details, amount of funds spent and more. For additional informational you can check our article about the adveriser’s account here.
Influencer`s personal account
Upon registration on the platform Influencers get access to the personal account which is designed to provide bloggers with all the materials necessary to participate in advertisers` campaigns and make creative native ads.
Influencer’s account has the following features:
- browsing and choosing advertising tasks
- checking tasks in progress
- checking tasks history
- managing connected social media accounts
- inviting friends via referral program
- depositing and withdrawing ADH
- specifying user settings
To start performing tasks an influencer takes these steps:
- register an account
- connect social media account (Instagram and/or YouTube)
- choose a task to execute
Influencer’s personal account features a CPM calculation model. It is a good step forward to make market benchmark that can be used by many influencers to get the fairest price for their publications.
It is important to know that a special influencer filter has been introduced. Only those influencers who have specified all the required information and meet the minimal subscribers requirements (500 for Instagram, 1000 for YouTube) can participate in advertising campaigns. These measures have been taken to prevent possible fraud.
It is the Artificial Intelligence modules which make the whole automation possible. It does the main job — selects bloggers to perform a task and recognizes photo, video and speech for task verification.
There are different types of Neural Networks (NN) under the hood, so let’s focus on them.
For recognition of certain objects we use convolutional neural network, CNN, which is as a standard in e-vision. It is a class of deep, feed-forward Neural Network working on custom algorithms optimised by our developers. It has been successfully applied to analyzing visual imagery.
Our AI module is 98% accurate identifying objects in a picture. This result is comparable to the results of such industry leaders as Facebook and Google. Meanwhile, the possibility of false positive finding constitutes approximately 10%. At the time AdHive developers are working on solutions to improve both indicators.
At the moment the AI can do the following to verify native advertisements:
- locate an object in a publication to be verified
- define the object`s boundaries with low bias (20–30 pixels)
- identify several different classes of objects in one publication, which allows to give bloggers complex tasks
- identify complex objects thanks to hierarchical database of objects and custom AI training tool (“TrainTool”)
To recognize objects in video, we use an optimized algorithm for high-speed video processing. Our tool can do a real-time object recognition and determine both number of object`s appearances in a video and amount of time an object stays in the frame, moreover it can also identify if a video contains other brands.
Here is a table of some popular real time object detection systems comparison table (mAP — mean average precision, which indicates the system`s accuracy, FPS — frames per second, which indicates the amount of frames the system can analyze in one second).
For sound recognition we use recurrent neural network, RNN, which works outstandingly well with backpropagation.
This latest generation speech recognition module was designed by AdHive developers to increase platform efficiency and enable end-to-end training while using a pre-trained language model. Upon training the module on a 20 hours speech sample we achieved 79,64% recognition accuracy.
The AI isn`t the only technical feature that makes the platform work. Apart from it there are other vital tools such as billing and scoring systems, statistics module and scaling mechanism.
This is a crucial part of the Platform which makes all the intra-platform transactions possible.
- Customized payment system — “payment-srv”, available as a docker container
- Smoke tests for the payment system
- Fully developed API
- JMX performance monitoring
Billing is vital for the needs of AdHive Platform. It is a sophisticated system which we are proud of and it deserves a separate article describing all the details. The article will be published on our medium in the near future.
The main purpose of Scoring is to estimate an advertising campaign CPM, which is calculated on the basis of set indicators through machine learning. This system, which takes its name from banking, makes Influencers ratings based on their channel characteristics. It considers different audience features such as average age, sex and location, amount of views and engagement rate. Such rating allows the AI to find Influencers who match advertiser`s task requirements the best.
This module is designed to use several calculation models. It allows to consider different pricing policy features existing in different regions and on different platforms. Scoring module`s architecture allows to connect other trained modules, which facilitates scaling up for new regions and social media.
Here is an illustration of the scoring mechanics
The system gets a request for CPM calculation and transfers it to the Conveyer. This module structures the input information and uploads it to the ComplexRegressor which picks a suitable pricing model and calculates the CPM.
To maintain the platform`s operation AdHive collects detailed information about influencer`s audience. Therefore, a special analytics module (STAT) was developed to ensure that bloggers publications` statistics are tracked and stored properly.
The STAT serves for 2 main purposes:
- Collecting data from social media
- Collecting users`s posts and accounts analytics
Since the AdHive mission is to facilitate native ads placement globally, it is important for the system to scale up easily.
It can be achieved by clustering the database and increasing the number of nodes of the service itself. The STAT module is divided into two sub-modules: social media parser and data analysis module. Such a divide makes it possible to tune the whole analytics module to scale up for different regions and social media easily.
The AI`s recognition accuracy depends on the amount of training materials and learning methods. For this purpose AdHive has developed “TrainTool”, special module aimed at facilitating the process of training our AI to recognize objects.
This tool allows to upload training samples for our recognition modules. Moreover, no special skill is required to train our AI.
Check our Github to get more information.
Further development according to our roadmap
Upon the Beta release we will focus on the following:
- Community member personal account
- Mobile app for influencers
- Influencers rating system
- Further AI development
- Advertising campaign results detailed dashboard
- Built-in ADH to fiat conversion and withdrawal through popular third party payment systems
We would like to invite influencers and advertisers to join AdHive upon the release and enjoy using the platform. Feel free to give your feedback, we appreciate it!