Pronto(a) para começar?
Execute campanhas de UA eficazes baseadas em ML (machine learning) com base na segmentação contextual.
Calculadora contextual
Com a depreciação iminente do IDFA no iOS 14, os anunciantes são desafiados a encontrar uma solução alternativa de segmentação que não seja baseada no usuário, mas que ainda gere CPIs competitivos e usuários valiosos. O jogo mudou, mas o objetivo continua o mesmo: crescimento eficaz.
Como a segmentação contextual agora assume uma importância central, para obter informações que vão além da categoria da loja, o que é de fácil manipulação, usamos um método de representação de palavras denominado ELMo para representar cada pacote (ou "bundle") como um vetor, permitindo a medição da "distância" contextual entre aplicativos. A distância calculada é então integrada às centenas de outros elementos dos nossos modelos de previsão e desempenha um papel significativo em nossa lógica de lances.
Para destacar sua eficácia, criamos a Calculadora de Distância Contextual. Ao utilizá-la, você verá quais são os aplicativos mais próximos do seu em termos de mecânica, tema, recursos e muito mais, com base em uma comparação detalhada entre cada texto de descrição na loja de aplicativos.
Encontre aplicativos contextualmente relacionados
Nome do aplicativo
Categoria
Principais aplicativos contextualmente relacionados
| No | App name | Confidence |
|---|---|---|
| 1 | Billion Cash Slots-Casino Game | 100% Confidence Level 100% - High Confidence As context is highly nuanced, we also feed our models with the level of confidence in the contextual relationship between both apps. Confidence is effected by the length of the app description, the quantity of ambiguous words in it, and overall clarity. |
| 2 | Eon Slots Casino Vegas Game | 90% Confidence Level 90% - High Confidence As context is highly nuanced, we also feed our models with the level of confidence in the contextual relationship between both apps. Confidence is effected by the length of the app description, the quantity of ambiguous words in it, and overall clarity. |
| 3 | Cash Tornado Slots - Casino | 50% Confidence Level 50% - Moderate Confidence As context is highly nuanced, we also feed our models with the level of confidence in the contextual relationship between both apps. Confidence is effected by the length of the app description, the quantity of ambiguous words in it, and overall clarity. |
| 4 | Super Vegas Slots Casino Games | 44% Confidence Level 44% - Moderate Confidence As context is highly nuanced, we also feed our models with the level of confidence in the contextual relationship between both apps. Confidence is effected by the length of the app description, the quantity of ambiguous words in it, and overall clarity. |
| 5 | Vegas Nights Slots | 40% Confidence Level 40% - Moderate Confidence As context is highly nuanced, we also feed our models with the level of confidence in the contextual relationship between both apps. Confidence is effected by the length of the app description, the quantity of ambiguous words in it, and overall clarity. |
| 6 | Tycoon Casino - Vegas Slots | 90% Confidence Level 90% - High Confidence As context is highly nuanced, we also feed our models with the level of confidence in the contextual relationship between both apps. Confidence is effected by the length of the app description, the quantity of ambiguous words in it, and overall clarity. |
| 7 | Winning Slots Las Vegas Casino | 90% Confidence Level 90% - High Confidence As context is highly nuanced, we also feed our models with the level of confidence in the contextual relationship between both apps. Confidence is effected by the length of the app description, the quantity of ambiguous words in it, and overall clarity. |
| 8 | Cash Frenzy - Slots Casino | 60% Confidence Level 60% - Moderate Confidence As context is highly nuanced, we also feed our models with the level of confidence in the contextual relationship between both apps. Confidence is effected by the length of the app description, the quantity of ambiguous words in it, and overall clarity. |
| 9 | Vegas Slots - 7Heart Casino | 10% Confidence Level 10% - Low Confidence As context is highly nuanced, we also feed our models with the level of confidence in the contextual relationship between both apps. Confidence is effected by the length of the app description, the quantity of ambiguous words in it, and overall clarity. |
| 10 | Slots Casino - Vegas Slot Machine Games | 90% Confidence Level 90% - High Confidence As context is highly nuanced, we also feed our models with the level of confidence in the contextual relationship between both apps. Confidence is effected by the length of the app description, the quantity of ambiguous words in it, and overall clarity. |
1
Billion Cash Slots-Casino Game
100%
Confidence Level
100% - High Confidence
As context is highly nuanced, we also feed our models with the level of confidence in the contextual relationship between both apps.
Confidence is effected by the length of the app description, the quantity of ambiguous words in it, and overall clarity.
2
Eon Slots Casino Vegas Game
90%
Confidence Level
90% - High Confidence
As context is highly nuanced, we also feed our models with the level of confidence in the contextual relationship between both apps.
Confidence is effected by the length of the app description, the quantity of ambiguous words in it, and overall clarity.
3
Cash Tornado Slots - Casino
50%
Confidence Level
50% - Moderate Confidence
As context is highly nuanced, we also feed our models with the level of confidence in the contextual relationship between both apps.
Confidence is effected by the length of the app description, the quantity of ambiguous words in it, and overall clarity.
4
Super Vegas Slots Casino Games
44%
Confidence Level
44% - Moderate Confidence
As context is highly nuanced, we also feed our models with the level of confidence in the contextual relationship between both apps.
Confidence is effected by the length of the app description, the quantity of ambiguous words in it, and overall clarity.
5
Vegas Nights Slots
40%
Confidence Level
40% - Moderate Confidence
As context is highly nuanced, we also feed our models with the level of confidence in the contextual relationship between both apps.
Confidence is effected by the length of the app description, the quantity of ambiguous words in it, and overall clarity.
6
Tycoon Casino - Vegas Slots
90%
Confidence Level
90% - High Confidence
As context is highly nuanced, we also feed our models with the level of confidence in the contextual relationship between both apps.
Confidence is effected by the length of the app description, the quantity of ambiguous words in it, and overall clarity.
7
Winning Slots Las Vegas Casino
90%
Confidence Level
90% - High Confidence
As context is highly nuanced, we also feed our models with the level of confidence in the contextual relationship between both apps.
Confidence is effected by the length of the app description, the quantity of ambiguous words in it, and overall clarity.
8
Cash Frenzy - Slots Casino
60%
Confidence Level
60% - Moderate Confidence
As context is highly nuanced, we also feed our models with the level of confidence in the contextual relationship between both apps.
Confidence is effected by the length of the app description, the quantity of ambiguous words in it, and overall clarity.
9
Vegas Slots - 7Heart Casino
10%
Confidence Level
10% - Low Confidence
As context is highly nuanced, we also feed our models with the level of confidence in the contextual relationship between both apps.
Confidence is effected by the length of the app description, the quantity of ambiguous words in it, and overall clarity.
10
Slots Casino - Vegas Slot Machine Games
90%
Confidence Level
90% - High Confidence
As context is highly nuanced, we also feed our models with the level of confidence in the contextual relationship between both apps.
Confidence is effected by the length of the app description, the quantity of ambiguous words in it, and overall clarity.
| No | App name | Confidence |
|---|---|---|
| 11 | Epic Vegas Slots - Casino Game | 40% Confidence Level 40% - Moderate Confidence As context is highly nuanced, we also feed our models with the level of confidence in the contextual relationship between both apps. Confidence is effected by the length of the app description, the quantity of ambiguous words in it, and overall clarity. |
| 12 | Slots of Vegas | 90% Confidence Level 90% - High Confidence As context is highly nuanced, we also feed our models with the level of confidence in the contextual relationship between both apps. Confidence is effected by the length of the app description, the quantity of ambiguous words in it, and overall clarity. |
| 13 | Slots: Vegas Casino Slot Games | 90% Confidence Level 90% - High Confidence As context is highly nuanced, we also feed our models with the level of confidence in the contextual relationship between both apps. Confidence is effected by the length of the app description, the quantity of ambiguous words in it, and overall clarity. |
| 14 | Vegas Casino & Slots: Slottist | 88% Confidence Level 88% - High Confidence As context is highly nuanced, we also feed our models with the level of confidence in the contextual relationship between both apps. Confidence is effected by the length of the app description, the quantity of ambiguous words in it, and overall clarity. |
| 15 | LuckyBomb Casino Slots | 90% Confidence Level 90% - High Confidence As context is highly nuanced, we also feed our models with the level of confidence in the contextual relationship between both apps. Confidence is effected by the length of the app description, the quantity of ambiguous words in it, and overall clarity. |
| 16 | Slots Mega Win Casino Game | 75% Confidence Level 75% - Moderate-High Confidence As context is highly nuanced, we also feed our models with the level of confidence in the contextual relationship between both apps. Confidence is effected by the length of the app description, the quantity of ambiguous words in it, and overall clarity. |
| 17 | High 5 Casino: Home of Slots | 90% Confidence Level 90% - High Confidence As context is highly nuanced, we also feed our models with the level of confidence in the contextual relationship between both apps. Confidence is effected by the length of the app description, the quantity of ambiguous words in it, and overall clarity. |
| 18 | Winner Slots Casino Games | 20% Confidence Level 20% - Low Confidence As context is highly nuanced, we also feed our models with the level of confidence in the contextual relationship between both apps. Confidence is effected by the length of the app description, the quantity of ambiguous words in it, and overall clarity. |
| 19 | Cash Tap Casino: Slot Machines | 90% Confidence Level 90% - High Confidence As context is highly nuanced, we also feed our models with the level of confidence in the contextual relationship between both apps. Confidence is effected by the length of the app description, the quantity of ambiguous words in it, and overall clarity. |
| 20 | Cash Tap Casino: Slot Machines | 90% Confidence Level 90% - High Confidence As context is highly nuanced, we also feed our models with the level of confidence in the contextual relationship between both apps. Confidence is effected by the length of the app description, the quantity of ambiguous words in it, and overall clarity. |
11
Epic Vegas Slots - Casino Game
40%
Confidence Level
40% - Moderate Confidence
As context is highly nuanced, we also feed our models with the level of confidence in the contextual relationship between both apps.
Confidence is effected by the length of the app description, the quantity of ambiguous words in it, and overall clarity.
12
Slots of Vegas
90%
Confidence Level
90% - High Confidence
As context is highly nuanced, we also feed our models with the level of confidence in the contextual relationship between both apps.
Confidence is effected by the length of the app description, the quantity of ambiguous words in it, and overall clarity.
13
Slots: Vegas Casino Slot Games
90%
Confidence Level
90% - High Confidence
As context is highly nuanced, we also feed our models with the level of confidence in the contextual relationship between both apps.
Confidence is effected by the length of the app description, the quantity of ambiguous words in it, and overall clarity.
14
Vegas Casino & Slots: Slottist
88%
Confidence Level
88% - High Confidence
As context is highly nuanced, we also feed our models with the level of confidence in the contextual relationship between both apps.
Confidence is effected by the length of the app description, the quantity of ambiguous words in it, and overall clarity.
15
LuckyBomb Casino Slots
90%
Confidence Level
90% - High Confidence
As context is highly nuanced, we also feed our models with the level of confidence in the contextual relationship between both apps.
Confidence is effected by the length of the app description, the quantity of ambiguous words in it, and overall clarity.
16
Slots Mega Win Casino Game
75%
Confidence Level
75% - Moderate-High Confidence
As context is highly nuanced, we also feed our models with the level of confidence in the contextual relationship between both apps.
Confidence is effected by the length of the app description, the quantity of ambiguous words in it, and overall clarity.
17
High 5 Casino: Home of Slots
90%
Confidence Level
90% - High Confidence
As context is highly nuanced, we also feed our models with the level of confidence in the contextual relationship between both apps.
Confidence is effected by the length of the app description, the quantity of ambiguous words in it, and overall clarity.
18
Winner Slots Casino Games
20%
Confidence Level
20% - Low Confidence
As context is highly nuanced, we also feed our models with the level of confidence in the contextual relationship between both apps.
Confidence is effected by the length of the app description, the quantity of ambiguous words in it, and overall clarity.
19
Cash Tap Casino: Slot Machines
90%
Confidence Level
90% - High Confidence
As context is highly nuanced, we also feed our models with the level of confidence in the contextual relationship between both apps.
Confidence is effected by the length of the app description, the quantity of ambiguous words in it, and overall clarity.
20
Cash Tap Casino: Slot Machines
90%
Confidence Level
90% - High Confidence
As context is highly nuanced, we also feed our models with the level of confidence in the contextual relationship between both apps.
Confidence is effected by the length of the app description, the quantity of ambiguous words in it, and overall clarity.
Resultados do pacote
Insira um ID do pacote e clique em "Calcular" para ver os aplicativos com o contexto mais similar
Por que ELMo?
Ao desenvolver os recursos de aproximação contextual para os nossos modelos de previsão de instalação e compra, consideramos inicialmente duas noções principais: 1) a de que o contexto desempenha um papel importante na previsão de comportamento do usuário e de engajamento com anúncios e 2) a de que esse contexto pode envolver diferenças sutis, mas significativas.
Com essas noções, o uso de categorias de lojas tornou-se inútil, uma vez que elas incluem uma parcialidade inerentee, em alguns casos, imprecisões notórias devido a questões de otimização ASO, ao passo que outras abordagens, como a categorização (clustering) usando métodos de modelagem de tópicos (como LDA ou W2V), não geraram o ganho de informações e a melhoria de precisão que buscávamos. Assim, percebemos que seria necessário encontraruma forma de representar cada pacote (bundle) individualmente.
Após avaliar a complexidade da tarefa e as ferramentas normalmente usadas para resolver problemas relacionados à linguagem com o machine learning, a pesquisa dos nossos cientistas de dados culminou no teste do ELMo. O ELMo faz uso de uma rede neural para aprender associações entre palavras e seu significado quando utilizadas em diferentes contextos, aprendendo em que frases elas são usadas dentro de um enorme conjunto de dados de 5,5 bilhões de tokens (palavras e seus elementos componentes).
Após a preparação do modelo ELMo, ele exibe uma representação de cada palavra como um vetor, o que nos permite medir a distância do cosseno entre cada palavra, indicando o nível de similaridade semântica e sintática entre elas. Uma vez que considerar cada palavra é fundamental para compreender o contexto, o ELMo se mostra uma das alternativas mais confiáveis para uma tarefa com tal complexidade.
Um exemplo de como os vetores de palavras são contextualmente próximos e, ao mesmo tempo, medidos de forma eficaz pela distância do cosseno
- Cluster 1
- Cluster 2
- Cluster 3
- Vetores de palavras
- Peso
- Representação do pacote
- Modelo de ML
Uma representação de como um vetor de pacote é formado a partir de uma agregação de soma ponderada de suas palavras
Representação vetorial do pacote
Depois de gerar os vetores de palavras, o próximo passo é agregá-los em cada pacote, de forma que seu contexto seja representado com eficácia. Ostensivamente, incorporando toda a descrição da loja em uma única representação vetorial, para que possamos medir a "distância" contextual entre diferentes aplicativos.
Essa etapa do nosso processo de pesquisa e desenvolvimento revelou-se a mais complicada. Simplesmente calcular a média dos vetores de todas as palavras da descrição do aplicativo na loja de aplicativos parece razoável, mas incorreria na perda de dados valiosos relacionados à frequência com que palavras específicas aparecem ou se repetem na descrição do aplicativo específico, bem como em todo o corpus.
Após um longo processo de testagem, a abordagem que gerou o maior ganho de informações nos testes foi baseada em uma estatística numérica básica em PLN (Processamento de Linguagem Natural, a área do machine learning que se concentra na análise de texto) – TF–IDF (sigla em inglês para "Term Frequency– Inverse Document Frequency", ou "Frequência de Termo – Frequência Inversa do Documento" em português) –, que basicamente indica a raridade de palavras específicas dentro de um corpus.
Por fim, a nossa representação de pacote evoluiu, tornando-se uma média ponderada de cada vetor de palavras, multiplicada por seu valor TF-IDF normalizado, a fim de enfatizar as palavras mais raras e matizadas que geralmente expressam temas, mecânicas e características únicas de cada aplicativo – fornecendo, assim, informações um tanto granulares, mas valiosas aos nossos modelos.