Links zu den Implementierungen neuronaler Konversationsmodelle für verschiedene Frameworks. Beiträge sind willkommen.
Theano, Lasagne
Ein Dialogsystem, das in der Lage ist, Emotionen in einer Textkonversation auszudrücken. Siehe Online-Demo.
Beispielergebnisse für neutrale Emotionen:
hi → hey you
how are you ? → i ' m fine , what about you ?
how old are you ? → i ' m 16 , how old are you ?
are you a leader or a follower ? → i ' m a man who is a politician .
my name is david . what is my name ? → i love your name .
is the sky blue or black ? → the sky is blue
you love anime . → i ' m not a weeb
Tensorflow
Schöne Ergebnisse:
Hi → Hi.
What is your name ? → Laura.
What does that mean ? → I dunno.
How old are you ? → thirty-five.
Who is Laura ? → My brother.
Say 'goodbye' → Alright.
Two plus two → manny...
Modifizierter Code des maschinellen Übersetzungsmodells. In diesem Code ist keine Antwort-Randomisierung implementiert, sodass die Modelle jedes Mal mit der gleichen Phrase antworten:
hello baby → hello
how old are you ? → twenty .
i am lonely → i am not
nice → you ' re not going to be okay .
so rude → i ' m sorry .
are you a robot or human? → no .
are you better than siri? → yes .
Fackel
Quellcode für eine Reihe superschöner Artikel von Jiwei Li – einer meiner Lieblingsforschungen im Bereich neuronaler Konversationssysteme. Einen Blick wert.
tell me ... how long have you had this falling sickness ? → a few months, I guess .
so i had the doctors test sammy ’s response to conditioning . → so he took the pills .
what are you going to do ? → i ’m going to get you a new car .
they fear your power your intellect . → i ’m not afraid of your power .
Hello? → Hi.
How are you? → I'm fine.
What's your name? → It's hard to describe.
How so? → I'm not sure.
What color is the sky? → It's blue.
What is your job? → It's not that i'm a fucking werewolf!
Keras
Die beliebteste Implementierung der Seq2seq-Architektur auf GitHub. Es gibt jedoch noch keine Hinweise darauf, dass mit diesem Repo gute Ergebnisse erzielt werden.
Ein Wrapper für farizrahman4u/seq2seq, der zum Ausführen von Experimenten verwendet wird. Bisher wurden keine guten Ergebnisse erzielt.
Ich muss das überprüfen.
Ein neuer generativer Chatbot, dessen Training in wenigen Epochen konvergiert, einschließlich eines Modells, das vorab auf einem kleinen, aber konsistenten Datensatz trainiert wurde, der aus Dialogen von Online-Englischkursen gesammelt wurde.
user: Hello! How are you?
computer: i am fine , ok , well , thank you . (with probability of 0.540239)
user: What's your name?
computer: janet . (with probability of 0.319482)
user: What do you like to do now?
computer: let me think a minute . (with probability of 0.286760)
user: I want to chat about computer science.
computer: i am glad to hear there . (with probability of 0.293063)
user: Are you a real girl or a computer program?
computer: i am self employed . i would love to be well . (with probability of 0.026870)
user: Why don't you look for a job?
computer: i do not know . i am the bank . (with probability of 0.053494)
user: So, you don't need a job!
computer: no . i have to make a lot of money . (with probability of 0.001279)
Ein toller Überblick über vorhandene Konversationsdatensätze.
Holen Sie sich viele rohe Filmuntertitel (~1,2 GB)
~ 40 MB nach Löschen der technischen Daten.
Lesen Sie diese großartige Rezension von Richard Krisztian Csaky: Chatbot and Related Research Paper Notes with Images
sowie zu seinem Berichtspapier im PDF-Format: Deep Learning Based Chatbot Models